Johan G. F. Belinfante, Bernard Kolman

A Survey of lie Groups
and Lie Algebras

with Applications and
Computational Methods
http://avaxho.me/blogs/ChrisRedfield
SIAM's Classics in Applied Mathematics series consists of books that were previousi
allowed to go out of print These books are republished by SIAM as a professional service
because they continue to be important resources for mathematical scientists.
Editor-in-Chief
Robert E O'Malley, Jr., University of Washington
Editorial Board
Richard A. Brualdi, University of Wisconsin-Madison
Herbert B. Keller, California Institute of Technology
Andrzej Z. Manitius, George Mason University
Ingram Olkin, Stanford University
Stanley Richardson, University of Edinburgh
Ferdinand Verhulst, Mathematisch Instituut, University of Utrecht
Classics in Applied Mathematics

C. C. Lin and L A. Segel, Mathematics Applied to Deterministic Problems in the Natural
Sciences
Johan G. F. Belinfante and Bernard Kolman, A Survey of Lie Groups and Lie Algebras with
Applications and Computational Methods
James M. Ortega, Numerical Analysis: A Second Course
Anthony V. Fiacco and Garth P. McCormick, Nonlinear Programming: Sequential
Unconstrained Minimization Techniques
F. H. Clarke, Optimization and Nonsmooth Analysis
George F. Carrier and Carl E. Pearson, Ordinary Differential Equations
Leo Breiman, Probability
R. Bellman and G. M. Wing, An Introduction to Invariant Imbedding
Abraham Berman and Robert J. Plemmons, Nonnegative Matrices in the
Mathematical Sciences
Olvi L Mangasarian, Nonlinear Programming
*Carl Friedrich Gauss, Theory of the Combination of Observations least Subject to Errors:
Part One, Part Two, Supplement Translated by G. W. Stewart
Richard Bellman, Introduction to Matrix Analysis
U. M. Ascher, R. M. M. Mattheij, and R. D. Russell, Numerical Solution of Boundary Value
Problems for Ordinary Differential Equations
K. E Brenan, S. L Campbell, and L R. Petzold, Numerical Solution of Initial-Value Problems
in Differential-Algebraic Equations
Charles L Lawson and Richard J. Hanson, Solving Least Squares Problems
J. E Dennis, Jr. and Robert B. Schnabel, Numerical Methods for Unconstrained Optimization
and Nonlinear Equations
Richard E Barlow and Frank Proschan, Mathematical Theory of Reliability
Cornelius Lanczos, Linear Differential Operators
Richard Bellman, Introduction to Matrix Analysis, Second Edition
Beresford N. Parlett, TTie Symmetric Eigenvalue Problem
Richard Haberman, Mathematical Models: Mechanical Vibrations, Population Dynamics, and
Traffic Flow
Peter W. M. John, Statistical Design and Analysis of Experiments
*'First time in print

A Survey of Lie Groups
and Lie Algebras
Johan G. F. Belinfante
Carnegie-Mellon University
Pittsburgh, Pennsylvania
Bernard Kolman
Drexel University
Philadelphia, Pennsylvania
asm..
Society for Industrial and Applied Mathematics
Philadelphia
Copyright 1972, 1989 by the Society for Industrial and Applied Mathematics.
This SIAM edition is an unabridged, corrected republication of the work first published by
SLAM in 1972.
10 9 8 7 6 5 4 3 2
The work of Bernard Kolman was supported in part by the Air Force Office of Scientific
Research, Air Force Systems Command, United States Air Force, under AFOSR Grant
69-1797.
All rights reserved. Printed in the United States of America. No part of this book may be
reproduced, stored, or transmitted in any manner without the written permission of the
publisher. For information, write the Society for Industrial and Applied Mathematics, 3600
University City Science Center, Philadelphia, PA 19104-2688.
Library of Congress Cataloging-in-Pdblication Data
Belinfante, Johan G. F.
A survey of Lie groups and Lie algebras with
applications and computational methods.
Includes bibliographical references.

1. Lie groups. 2. Lie algebras. I. Kolman,
Bernard, 1932- . II. Tide.
QA387.B43 1989 512'.55 89-19699
ISBN 0-89871-243-2
SL3J1L is a registered trademark.

Preface to the Classic Edition
The validity of the first paragraph of the Preface to the original edition
Applications of the theory of Lie groups, Lie algebras and their
representations are many and varied. This is a rapidly growing
field through which one can bring to bear many powerful meth-
ods of modern mathematics.
written 18 years ago, has been borne out by the profusion of published
work in this area.
During this period of time we have seen the appearance of several
other presentations of Lie theory directed to various special fields. Another
trend has been the increased emphasis on computational Lie algebra, with
heavy reliance on computers.
We have naturally been gratified and much encouraged by the warm
reception of the first edition. When it became known that a classic edition
was to be published, we received many valuable suggestions for additional
topics to be included. We regret that most of these could not be readily
incorporated without completely changing the outlook, and greatly in-
creasing the size, of the book. We have corrected a few minor misprints
which appeared in the original edition, but the character of the book, espe-
cially its focus on the classical representation theory and its computa-
tional aspects, has not changed.
We thank the many people who wrote us with their comments and
suggestions. We also thank the entire staff of SIAM for their interest and
unfailing cooperation during all phases of this project.
This page intentionally left blank
Acknowledgments
We express our thanks to Harvey A. Smith who collaborated with us

on the first two SLAM Review papers which served as a nucleus for this
monograph; to Vishnu K. Agrawala and to Robert E. Beck who collabo-
rated with us on the computer approaches to representation theory which
are described in this book; to E Cartier for two discussions on the termi-
nology concerning layers and levels and on computing aspects of represen-
tation theory; and to T. Shimpuku, who furnished us with valuable informa-
tion on Clifford algebras. We are grateful to many readers of the SLAM
Review survey papers for numerous valuable suggestions, some of which
have been incorporated in the present work. Robert E. Beck and Roger W.
Brockett deserve our thanks for reading the manuscript and offering many
suggestions and criticisms.
We are grateful to Miss Susan R. Gershuni who typed the entire manu-
script and its revisions and to our wives, Irene and Lillie, who helped with
the preparation of the manuscript.
Contents
Introduction 3
Chapter 1 Lie Groups and Lie Algebras 5
1.1 The general linear group 5
1.2 Orthogonal and unitary groups 6
1.3 Groups in geometry 7
1.4 The exponential mapping 11
1.5 Lie and associative algebras 12
1.6 Lie groups 13
1.7 Lie algebras of Lie groups 15
1.8 Vector fields 18
1.9 Lie theory of one-parameter groups 19
1.10 Matrix Lie groups 21
1.11 Poisson brackets 24
1.12 Quantum symmetries 26
1.13 Harmonic oscillators 30
1.14 Lie subgroups and analytic homomorphisms 31
1.15 Connected Lie groups 32
1.16 Abelian Lie groups 34
1.17 Low-dimensional Lie groups 35
1.18 The covering group of the rotation group 36
1.19 Tensor product of vector spaces 38
1.20 Direct sums of vector spaces 42
1.21 The lattice of ideals of a Lie algebra 43
1.22 The Levi decomposition of a Lie algebra 44
1.23 Semisimple Lie algebras 45
1.24 The Baker-Campbell-Hausdorff formula 47
Chapter 2 Representation Theory 51
2.1 Lie group representations 51
2.2 Modules over Lie algebras 53
2.3 Direct sum decompositions of Lie modules 56
2.4 Lie module tensor product 57
2.5 Tensor and exterior algebras 59
2.6 The universal enveloping algebra of a Lie algebra 63
2.7 Nilpotent and Cartan subalgebras 65
2.8 Weight submodules 66
2.9 Roots of semisimple Lie algebras 67
2.10 The factorization method and special functions 70
x Contents
2.11 The Cartan matrix 73
2.12 The Weyl group 74
2.13 Dynkin diagrams 76
2.14 Identification of simple Lie algebras 78
2.15 Construction of the Lie algebra A2 79
2.16 Complexification and real forms 80
2.17 Real forms of the Ue algebra Al 83
2.18 Angular momentum theory 86
Chapter 3 Constructive Methods 91
3.1 Raising and lowering subalgebras 91
3.2 Dynkin indices 93
3.3 Irreducible representations of Al 95
3.4 The Casimir subalgebra 97
3.5 Irreducible representations of A2 99
3.6 Characters 101
3.7 Computation of the Killing form 103
3.8 Dynkin's algorithm for the weight system 106
3.9 Freudenthal's algorithm 109
3.10 The Weyl character formula 111
3.11 The Weyl dimension formula 114
3.12 Characters of modules over the algebra A2 116
3.13 The Kostant and Racah character formulas 117
3.14 The Steinberg and Racah formulas for Clebsch-Gordan
series 119
3.15 Tensor analysis 122
3.16 Young tableaux 124
3.17 Contractions 128
3.18 Spinor analysis and Clifford algebras 131
3.19 Tensor operators 137
3.20 Charge algebras 141
3.21 Clebsch-Gordan coefficients 145
Bibliography 149
Index 159
A Survey of
Lie Groups and Lie Algebras
INTRODUCTION
The theory of Lie groups and Lie algebras is an area of mathematics in

which we can see a harmonious interaction between the methods of classical
analysis and modern algebra. This theory, a direct outgrowth of a central
problem in the calculus, has today become a synthesis of many separate
disciplines, each of which has left its own mark.
The theory of Lie groups was developed by the Norwegian mathe-
matician Sophus Lie in the late nineteenth century in connection with his
work on systems of differential equations. Lie groups arise in the study of
solutions of differential equations just as finite groups arise in the study of
algebraic equations. The possibility of solving many differential equations
which arise in practical problems can often be traced to some geometrical
or other symmetry property of the problem. Indeed, Lie groups play a
fundamental role in many parts of geometry itself, and have contributed
significantly to the development of that subject. Thus, for many years Lie
theory has been found useful in such areas of pure mathematics as the
differential geometry of symmetric spaces. Conversely, geometrical methods
have played a dominant role in the structure theory of Lie groups and
Lie algebras.
In our exposition of this structure theory, we focus on the classification
of real and complex semisimple Lie algebras. We found this to be an excellent
place to introduce and apply many of the tools of classical linear algebra and
vector space theory, as well as the modern theory of modules.
The importance of Lie algebras and Lie groups for applied mathematics
and for physics has also become increasingly evident in recent years. In
applied mathematics, Lie theory remains a powerful tool for studying
differential equations, special functions and perturbation theory. Lie theory
often enters into physics either through the presence of exact kinematical
symmetries or through the use of idealized dynamical models having greater
symmetry than is present in the real world. These exact kinematical sym-
metries include rotational, translational and Galilean or Lorentz invariance,
as well as symmetries arising from the use of canonical formalism in both
classical and quantum mechanics. Broken symmetries arising from approxi-
mate models are encountered in atomic and nuclear spectroscopy, and in
elementary particle physics.
3
4 Introduction
Since much of the applied literature nowadays presumes a quite detailed

familiarity with the methods of representation theory, we include a leisurely
treatment of this crucial aspect of the subject. We discuss here the concepts of
representations and modules, the standard module operations, and the
classification of the finite-dimensional modules over semisimple Lie algebras.
We also discuss in detail various computational methods, including the
explicit construction of the irreducible modules, which are needed for
calculating Clebsch-Gordan and Racah coefficients.
The study of the constructive aspects of Lie module theory has become
practical recently through the use of computers. Computers provide a
valuable new research tool in Lie theory, which makes it possible to test
various theoretical conjectures about the representations of Lie algebras.
We believe this area of Lie theory will continue to grow rapidly for some time
to come. Our discussion of the use of computers, with the resulting emphasis
on constructive methods, is just one of the ways we have kept Lie theory
from becoming too abstract. In addition, we have throughout this volume
presented diverse applications to topics in mathematics and physics to
illustrate various points of Lie theory. In our selection of such topics we have
tried to indicate a few of the many interesting directions in which applied
research on the Lie theory has been carried out.
However, our primary aim has always been to use applications only to
provide some motivation for discussing various aspects of Lie theory itself.
We do not claim to have given a comprehensive treatment of even the major
applications. Lie theory finds application not only in elementary particle
physics and nuclear physics, but also in such diverse fields as continuum
mechanics, solid-state physics, cosmology and control theory. Much was
left out to keep the discussion within manageable dimensions, and the reader
will realize that to pursue the applications seriously, one must go beyond the
material presented here. Hopefully the bibliography will prove to be useful
as a guide for collateral reading, but it is necessarily incomplete and we must
apologize to those authors whose work has been left out.
CHAPTER 1
LIE GROUPS AND LIE ALGEBRAS
1.1 THE GENERAL LINEAR GROUP
Since many of the central ideas of Lie theory arose in the study of
geometry and linear algebra, it is fitting to begin with a review of some topics
in vector space theory so that we can begin to talk about some of the essential
tools, such as the Lie algebra and tensor algebra constructions. It is also
instructive to examine the classical linear Lie groups before getting involved
with the rather technical general definition of a Lie group.
We recall that a linear mapping a: V^ - V2 from a vector space Vl into a
vector space V2 is a mapping which preserves vector addition and scalar
multiplication:
and
where and A is a scalar.

In particular, a linear operator on a vector space V is a linear mapping
from Kinto itself. A linear operator has an inverse if and only if it is both one-
to-one and onto as a mapping. The set of all invertible linear operators on a
vector space V is a group called the general linear group on V.
In the general linear group we take the group operation to be the ordinary
composition of linear operators. A linear operator on a finite-dimensional
vector space is invertible if and only if its determinant is nonzero. The general
linear group on a finite-dimensional vector space consists therefore of all
linear operators with nonzero determinants. The set of all linear operators
with determinant one forms a subgroup called the special linear group on K
If we introduce a basis into the vector space, each linear operator can be
represented by a matrix and we can talk in terms of matrix groups. Thus, the
matrix of a linear operator a with respect to a basis e } , , en for Vis the array
of scalars [a'V] determined by
The general and special linear groups may then be regarded as groups of
matrices, the group operation being matrix multiplication. These groups of
5
6 Lie Groups and Lie Algebras 1.2
matrices can be defined for any field of scalars such as, for instance, the real
number field R or the complex number field C. Thus, in particular we obtain
the group of all nonsingular complex n x n matrices, called the complex
general linear group GL(n, C).
The real general linear group GL(n, R), consisting of all real nonsingular
n x n matrices, may be regarded as a subgroup of the complex general linear
group. The complex special linear group SL(n, C) is the subgroup of GL(n, C)
consisting of matrices with determinant one. The intersection of these two
subgroups is the real special linear group,
The special linear groups are sometimes also called unimodular groups. The
reader should be warned however that in abstract harmonic analysis the
same term, unimodular group, is used in a completely different sense [152].
1.2 ORTHOGONAL AND UNITARY GROUPS
Other important subgroups of the general linear group GL(n, C) are

familiar from elementary matrix theory. These include the group of complex
orthogonal matrices O(n, C) and the group of unitary matrices U(ri). We
recall that a matrix is orthogonal if its transpose is its inverse, while a matrix is
unitary if its transposed complex conjugate is its inverse. We obtain "special"
versions of these groups by taking their intersections with the complex special
linear group. The special complex orthogonal group is
The special unitary group is
The determinant of an orthogonal matrix can only be + 1 or 1. This is

because the determinant of the transpose of a matrix is the same as that of the
matrix, while the determinant of the inverse matrix is the reciprocal of its
determinant. Orthogonal matrices are classified as proper or improper,
depending upon whether their determinants are + 1 or 1. Thus we may also
describe the special orthogonal group as the group of all proper orthogonal
matrices.
By considering further intersections of the above groups with the real
general linear group, we obtain some groups which are important in
Euclidean geometry. The real orthogonal group is defined as
1.3 Groups in Geometry 7
The real orthogonal group may be considered as the group of all rigid motions
which leave a given point fixed. The rotation group is the special or proper
real orthogonal group,
The geometrical interpretation of the distinction between proper and

improper orthogonal matrices can be studied explicitly for low dimension.
For n = 2, every improper orthogonal matrix can be interpreted geometrically
as a reflection, but this is not the case for n = 3. For example, the parity or
inversion operation, represented by the negative of the unit matrix, is not a
mirror reflection for n = 3, but it is nevertheless improper. In general the
parity operation behaves differently in even and odd dimension, being a
rotation for even n and improper only for odd n. Thus, it is not too surprising
that the relation between the groups 0(n, R) and S0(n, R) is somewhat
different for even and odd n. For odd n, the group 0(n, R) is simply the direct
product of S0(n, R) with the group Z2 consisting of 1. This is not true for
even n as can be seen in the case n = 2. Indeed, since S0(2, R) and Z2
are Abelian, so is their direct product, while the group 0(2, R) is non-
Abelian. Nevertheless, the general relation between 0(n, R) and S0(n, R) is
rather simple and can be described formally in terms of quotient groups.
The mapping which assigns to each matrix its determinant provides a
homomorphism from the group 0(n, R) onto the finite group Z 2 . The kernel
of this homomorphism is just the rotation group S0(n, R), so that the
rotation group is a normal subgroup of 0(n, R). Hence, by the fundamental
homomorphism theorem, the quotient group of 0(n, R) with respect to
S0(n, R) is isomorphic to Z 2 .
1.3 GROUPS IN GEOMETRY
Matrix Lie groups also arise in various types of non-Euclidean geometry,

and are sometimes encountered in applications. An important example is the
Lorentz group, which is an indefinite-metric version of the orthogonal group
for the geometry of space-time in the special theory of relativity. Another
group, the symplectic group, plays an important role in the geometry of phase
space in classical mechanics. To obtain a deeper understanding of the
geometrical significance of the orthogonal and symplectic groups, we turn to
some elementary notions from the theory of bilinear forms [134].
A bilinear form on a vector space V over a field F is a function which
assigns to each ordered pair of vectors x and y in V a scalar (x, y) e IF, and
which satisfies the following properties for any scalars a and /? and any
vectors x, y and z in V:
A bilinear form (x, y) can be completely described by its matrix /?,-_,-

= (ei, 6j) with respect to any basis e^ , , en. Indeed, if we are given this
matrix, then for any vectors
we can compute
A bilinear form (x, y) is symmetric if (x, y) = (y, x) for all x and y in V.

If (x, y) is a symmetric bilinear form, then its matrix /? is equal to its own trans-
pose. Similarly, we say that a bilinear form (x, y) is antisymmetric if (x, y)
= (y, x) for all x and >> in V, and in this case /? is equal to the negative of its
transpose. A familiar example of a symmetric bilinear form is, of course, the
inner product in Euclidean vector geometry. In Riemannian geometry, the
metric tensor and the curvature tensor are both examples of symmetric
bilinear forms.
Symmetric bilinear forms also occur naturally in the study of linear
operators in vector spaces of finite dimension. We may define such a bilinear
form by using the trace operation TrK ( ) for linear operators on a vector
space V. If we introduce a basis, the trace of a linear operator is the sum of its
diagonal matrix elements
Since the trace of an operator is also equal to the sum of its eigenvalues,
counting multiple roots of the characteristic equation, it does not matter
which basis we use here. The set linF (K) of all linear operators on an n-di-
mensional vector space Fover a field F is itself an 2-dimensional vector space
over F. Then TrK(a/?) defines a bilinear form on the vector space lin F (K).
This type of symmetric bilinear form is important in the study of the structure
of Lie groups and Lie algebras.
The introduction of a symmetric or antisymmetric bilinear form in a
vector space leads naturally to the idea of orthogonality of vectors. Vectors
x and y in V are said to be orthogonal if (x, y) = 0. Much of the theory of
symmetric bilinear forms also applies to antisymmetric bilinear forms, and
we may discuss the two cases simultaneously.
For a symmetric or antisymmetric bilinear form, the orthogonality of
vectors is a symmetric relation ; if x is orthogonal to y, then y is also ortho-
gonal to x. A nonzero vector x is said to be isotropic, or a null vector, if it is
orthogonal to itself, (x, x) = 0, and anisotropic if (x, x) ^ 0. Isotropic vectors
are of course not present in Euclidean geometry, but they often occur in more
general situations. In special relativity theory, isotropic vectors are called
light-like because light rays travel in straight lines along the directions of such
vectors. For symmetric bilinear forms over the complex number field,
1 .3 Groups in Geometry 9
isotropic vectors occur in any vector space having dimension ^2. If we are
dealing with antisymmetric bilinear forms over the real or complex number
field, then every vector is isotropic.
For the discussion of the orthogonal and symplectic groups, we shall
restrict our attention to nonsingular bilinear forms. A bilinear form is
nonsingular if the only vector orthogonal to every other vector is zero, that is,
if (x, y) = 0 for all y in V implies that x = 0. The bilinear form (x, y) is non-
singular if and only if its matrix ft is invertible, which means that det ft ^ 0.
It should be noted that whether or not a bilinear form is nonsingular, its
restriction to a subspace need not be nonsingular. Thus the notion of non-
singularity also leads to a classification of the subspaces of a vector space, and
this is often useful in the structure theory of vector spaces equipped with
bilinear forms.
An orthogonal linear operator a with respect to a nonsingular symmetric
bilinear form is a linear operator for which
(ax, ay) = (x,y)
holds for all x and y in K If x belongs to the kernel (null space) of an orthog-
onal linear operator a, then we have ax = 0, and hence for all y we find
(x, y) = (ax, ay) = 0. Since the bilinear form is nonsingular, x = 0 and a is
one-to-one, and if V is finite-dimensional, this further implies that a is
invertible. Thus the set of orthogonal linear operators in this case forms a
group, which we call the orthogonal group with respect to the given symmetric
bilinear form.
We obtain the matrix groups mentioned earlier by introducing bases for
vector spaces equipped with nonsingular bilinear forms. The matrix ft
describing the bilinear form depends on the choice of the basis ev , , en.
If ft is the matrix of this same bilinear form with respect to another basis
e\, , e'n, where e\ = Xj^Vj' then ft' is related to ft by the congruence
transformation
We can make use of such congruence transformations to bring the matrix ft

to a canonical form. It is a standard result of matrix theory that we can choose
the matrix a so that ft' is a diagonal matrix [191]. The exact form of the
diagonal matrix depends on whether we are considering bilinear forms over
the real or complex number fields.
In the complex case, when det ft ^ 0, we can arrange it so that ft' is the
identity matrix. Thus for a nonsingular symmetric bilinear form over the
complex number field, we can always find an orthonormal basis for V. A
basis e , , e is said to be orthonormal if
where the Kronecker delta <5i; is equal to 1 if i = j and 0 if i ^ j. The matrix of

an orthogonal linear operator with respect to an orthonormal basis satisfies
the condition that its inverse is its transpose. Therefore, we obtain just one
orthogonal group for each dimension in the complex case, namely the group
0(ii, C).
In the real case, if det ft ^ 0, we can find a matrix a so that /?' is diagonal
and its diagonal elements are 1. The signature (p, q), where p and q are the
numbers of + 1's and 1's respectively on the diagonal, is an invariant of the
real symmetric bilinear form. In the special case where p = n and q = 0, the
real symmetric bilinear form is said to be positive definite because (x, x) > 0
for all x ^ 0, and we obtain the ordinary real orthogonal group O(n, R). In
the general case, the group O(p, q ; R) which we obtain may be described as an
orthogonal group with an indefinite metric of signature (p, q) in a space of
dimension n = p + q. An example of this is the Lorentz group 0(3, 1 ; R),
which is just the orthogonal group of the Minkowski space-time R 4
(+ + + -)
The definition of the symplectic groups is somewhat similar to that of the
orthogonal groups. Just as the orthogonal groups consist of linear operators
which leave invariant a symmetric nonsingular bilinear form, the symplectic
groups consist of linear operators which leave invariant an antisymmetric
nonsingular bilinear form. A symplectic linear operator a is an operator
satisfying
for an antisymmetric nonsingular bilinear form. Nonsingular antisymmetric

bilinear forms can only occur in vector spaces V which have even dimension,
dim V = 2n. In any such space we can find a hyperbolic basis el, , en
such that
for all 1,7 = ! , - , .

To describe the matrices of symplectic linear operators with respect to
such a basis, we introduce the 2n x 2n matrix
where 1 denotes the n x n identity matrix. The matrix group Sp(n, C) is

then the set of all complex 2n x 2n matrices A which satisfy
where AT denotes the transpose of the matrix A. Closely related to this

complex symplectic group Sp(n, C) are some other matrix groups obtained by
taking intersections. Since we can find hyperbolic bases in both real and
complex vector spaces, there is also only one real symplectic group. This
real symplectic group is the intersection
1.4 The Exponential Mapping 11
We may also mention here the unitary symplectic group,
Taking intersections with SL(n, C) gives nothing new since every symplectic
linear operator already has determinant one [10]. The symplectic groups have
been applied to various problems in nuclear physics [121]. There are many
other matrix groups, and more extensive catalogues of them can be found
in the literature [120].
1.4 THE EXPONENTIAL MAPPING

The classical matrix groups which have been discussed so far are all
defined by imposing various algebraic conditions on the matrices belonging
to the group. When these conditions are written out in terms of the matrix
elements, we obtain complicated nonlinear equations to solve. It is remarkable
that in each case these nonlinear conditions can be replaced by an equivalent
set of linear conditions. This is in fact the essential idea behind the passage
from Lie groups to Lie algebras. The actual relation between Lie groups and
Lie algebras can be described in several different ways. To expedite matters,
we shall consider here the fastest way to get to the Lie algebra by using the
concept of matrix exponentiation. Although this method does not work for all
Lie groups, it is sufficient for the classical matrix groups.
For any square matrix A, we define exp (A) by the power series
where 7 is the identity matrix, and each matrix element converges

absolutely. Thus, for example, we have exp (0) = /. From the inverse
function theorem, it is easily seen that every matrix in some neighborhood
of the identity in GL(n, C) can be expressed as exp (A} for some n x n complex
matrix A. For GL(n, R), we can restrict the matrix A to be real. In the case of
GL(n, R), the exponential map does not yield the whole group, as can be seen
by considering the case where n = 1. The group GL(1, R) is the real line with
the origin removed, while the image of the exponential map is only the
positive real axis. This is the price that must be paid for using the exponential
map; that is, we may sometimes lose the global structure of the group. The
local structure however is completely preserved.
To assure that the matrix exp (A) will belong to a given one of the
classical matrix Lie groups we have discussed, we require the matrix A to
satisfy a suitable linear condition. For the special linear groups, we can
replace the nonlinear condition on exp (A) involving the determinant by a
linear condition on A involving the trace. If /. is an eigenvalue of A, then
exp (A) is an eigenvalue of exp (A) with the same eigenspace. The determinant
of exp (A), being the product of its eigenvalues, is equal to exp (Tr (A)) since
the trace Tr (A) is the sum of the eigenvalues of A. Hence the determinant of
exp (A) is one if the trace of A is zero.
We may also use the exponential map to get linear conditions in the
case of the orthogonal, unitary and symplectic groups. The transpose of the
exponential of a matrix is the exponential of its transpose, and the inverse of
the exponential of a matrix is the exponential of its negative. Hence, a
sufficient condition for exp (A) to be orthogonal is that A be antisymmetric.
Similarly, the complex conjugate of exp (A) is the exponential of the complex
conjugate of A. Hence, exp (A) is unitary when A is equal to the negative of its
transposed complex conjugate. Such a matrix is called anti-Hermitian.
Finally, we note that exp (^4) is symplectic if A satisfies ATJ + JA = 0, which
is again a linear condition.
In summary, the exponential mapping allows us to replace the classical
matrix Lie groups by sets of matrices which satisfy various linear conditions.
Such sets of matrices are of course closed under taking arbitrary real linear
combinations, and we may view these sets as vector spaces. These sets of
matrices are not closed under ordinary matrix multiplication. However, if A
and B are matrices which are, respectively, antisymmetric, anti-Hermitian or
traceless, then AB BA also has that property. Hence, the sets of matrices
with which one has to deal are not only vector spaces, but are also closed under
an additional operation, and one says that they form a Lie algebra. The study
of these Lie algebras, as well as others just like them, forms the central core
of the Lie theory.
1.5 LIE AND ASSOCIATIVE ALGEBRAS
Before studying the Lie algebras of the classical matrix groups in detail,
it is useful to discuss Lie algebras in general from a slightly more formal view-
point. In addition, we wish to discuss some relations existing between Lie
algebras and associative algebras. In general, any vector space endowed with
a bilinear vector multiplication is called a nonassociative algebra [207]. That
is, a nonassociative algebra consists of a vector space V together with any
bilinear mapping V x V -> V. The product of two vectors x, y in V then is
another vector xy in V. If the vector multiplication is associative, so that
(xy)z = x(yz), then we speak of an associative algebra. For example, the set
Imp (V) of all linear operators on a vector space Kover a field F is an associa-
tive algebra under the usual multiplication of linear operators. This same
vector space lin F (F) also has another algebra structure defined on it by
taking the product of linear operators a and /? to be the commutator ctfi /fe.
To distinguish this product from the usual product, we use a special square
bracket notation [a,)3] for it. The resulting algebra is a nonassociative
algebra of a certain special type called a Lie algebra.
Abstractly, a Lie algebra L is a vector space equipped with a product
[x, y] satisfying certain axioms [73], [82], [83], [135], [228]. We shall continue
to use the bracket notation for products when we deal with any Lie algebra.
One of the axioms for a Lie algebra is that the product [x, y] be bilinear, that
1.6 Lie Groups 13
is, linear in x and y separately. We also assume that the Lie product is-anti-
commutative,
Finally we assume that the Jacobi identity
holds for all vectors x, y, z in the Lie algebra. The Jacobi identity plays for
Lie algebras the same role that the associative law plays for associative
algebras. While we can define Lie algebras over any field, we shall in practice
only consider Lie algebras over the real and complex fields.
Undoubtedly the most familiar abstract Lie algebra is the real three-
dimensional vector space with the vector cross product as multiplication.
Obviously the vector cross product Lie algebra can also be defined for a
three-dimensional vector space over any field. Anticommutativity of the
vector cross product is familiar, and the Jacobi identity for cross products
can be verified by a direct computation, using
For Imp (V) with the commutator multiplication it is likewise easy to verify
directly that the Lie algebra axioms are satisfied. The procedure used to give
linF (V) a Lie algebra structure can be extended to any associative algebra. A
Lie product can be defined in any associative algebra by the commutator
[x, y] = xy yx, making it a Lie algebra. Here the anticommutativity axiom
is clearly satisfied, and the reader can check that in this case the Jacobi identity
for the commutator follows from the associative law for the ordinary product.
There is another and perhaps more revealing way in which Lie algebras
arise in the study of algebras. A derivation dot a nonassociative algebra A is a
linear operator on A satisfying the formal analogue of the Leibniz rule for
differentiating a product,
for all elements x and y in A. If p(x) is any polynomial in x, then the differential
operator p(x) d/dx is an example of a derivation of the algebra of polynomials
in x. In fact, the concept of a derivation is just an abstraction of the idea of a
first order differential operator. The set of all derivations on a nonassociative
algebra A is clearly a subspace of the linear operator algebra linF (A).
Although the product of derivations is in general not a derivation, the
commutator d^d2 - d2dl of two derivations is again a derivation. Thus the
set of all derivations of a nonassociative algebra is a Lie algebra, which we
call the derivation algebra of the given nonassociative algebra.
1.6 LIE GROUPS
The general theory of Lie groups seeks to unify and to extend the dis-
cussion of the various classical groups. The extension of the theory comes
about by treating not only matrix groups, but other types of groups as well.
Basically, a Lie group is the structure which naturally results when analytic
machinery is coupled with abstract group theory. Curvilinear coordinates,
derivatives, and power series become available as tools for the study of the
resulting structure. Technically, a Lie group is first of all required to be a
topological group, that is, it must be possible to assign a topology under which
the group operation and taking inverses are both continuous functions
[129], [164], [192]. The general concept of a topological group already leads
to a very rich theory in which some of the methods of calculus, especially
integration theory, can be applied. To make use of the full power of all the
methods of calculus, including differentiation, we must introduce an analytic
structure as well as a topology. A proper discussion of the analytic structure
of a Lie group requires the concept of an analytic manifold [38], [42], [54],
[64].
Historically, the concept of an analytic manifold grew out of the theory
of Riemann surfaces in complex variable theory [238]. The theory of differen-
tiate and analytic manifolds has since become the basis for modern
differential geometry [127], [230], The idea of defining a Lie group in terms
of analytic manifolds did not gain widespread acceptance until almost half
a century after Lie's original work [163]. Moreover, it turned out that any
connected real analytic manifold may always be analytically embedded in a
Euclidean space, so that it appears after all that manifold theory is somewhat
of a luxury [171]. Therefore, instead of using the intrinsic definition, we shall
regard a connected analytic manifold more intuitively as a smooth surface in a
Euclidean space of suitable dimensions. In general, an analytic manifold will
consist of several pieces if it is not connected. About each point of the
manifold there is required to be an open set with all points within this open
set being located by a curvilinear coordinate system. It is also required that
the assignment of coordinates to points be a continuous mapping with a
continuous inverse. Such an open subset of a manifold is called a coordinate
neighborhood or coordinate patch, and the coordinate system is said to give a
chart of this neighborhood. To avoid singular points like the poles on a
Mercator projection map, we use many charts, forming an atlas which covers
the manifold, each point of the manifold belonging to at least one chart. In
general, these charts may overlap somewhat so that two coordinate neighbor-
hoods may have a point in common. Overlapping charts are required to be
analytically related, meaning that for any common point, the coordinates in
each system are analytic functions of those in the other. A function defined
in Euclidean space R" is analytic at a point if it can be expressed as a con-
vergent Taylor series in some neighborhood of the point.
A Lie group is a group which is also an analytic manifold and for which
the group operations are analytic functions. Because of the analytic structure,
each element of the group is specified by some curvilinear coordinate system.
The multiplication of elements of a Lie group is required to be analytic in
terms of these coordinates. Thus if x and y are points of the group, and if
z = x o yt where o denotes the group operation, then we require that the
1.7 Lie Algebras of Lie Groups 15
coordinates of z be analytic functions of those of x and y. We need not

separately postulate that the mapping xi-^x" 1 be analytic, because this
actually follows from the analyticity of the multiplication [64] . The analyticity
requirements in the definition of a Lie group can be weakened to differen-
tiability requirements, since the latter imply the former in this case. As a
matter of fact, it is possible to characterize Lie groups by even weaker
conditions [170].
For completeness we should mention that the theory of Lie groups does
not provide the only possible way of unifying the classical groups. The theory
of algebraic groups is an alternative approach which has the advantage over
the Lie theory in that the restriction to the real or complex number fields can
be removed. Unfortunately, the theory becomes much more complicated
when general fields are permitted, and this more general concept has not yet
proved to be useful outside of pure mathematics.
1.7 LIE ALGEBRAS OF LIE GROUPS
The whole atlas of charts of a Lie group can be generated from the
charts at the identity element. If U is a coordinate neighborhood of an
element g, then the set g-1U obtained by letting g" 1 act on all elements of
U will contain a coordinate neighborhood of the identity element. By this
procedure, every coordinate neighborhood can be transported back to the
identity, and it suffices to study what happens close to the identity. For
convenience, coordinates in a neighborhood of the identity will be chosen
so that the coordinates of the identity are zero. If x is a member of the group,
we denote its coordinates by x'. If x <> y = 2, we have z'(x! , , x", y 1 , - - , / ) ,
by hypothesis, an analytic function which can be expanded in a convergent
Taylor series about the origin. Thus,
If the coordinate neighborhood is taken sufficiently small, the terms indicated

by dots become arbitrarily small with respect to the terms written out. If we
set both x and y equal to the identity element, then z will also be the identity
element, so that z'(0, 0) = 0. If we set y equal to the identity element, but leave
x arbitrary, this must reduce to the equation z = x. Hence, we must have
where Slj is the Kronecker delta and
A similar argument can be given by setting x equal to the identity, leaving y

arbitrary. We are then left with the series
It seems therefore that the local structure of a Lie group is determined by

the tensor a'jk. Actually, only the antisymmetric part of this tensor is relevant
because its symmetric part can be transformed away simply by introducing a
new system of coordinates. Indeed, we need only set
In these coordinates, group multiplication is given locally by
For convenience we may assume that such a coordinate transformation has

already been performed beforehand, if necessary, and we henceforth drop
the primes. It is not possible by means of further coordinate transformations
to eliminate the remaining structure tensor
It is of interest to study what group operation corresponds to this

structure tensor. Our formula shows that in lowest order, group multiplica-
tion corresponds simply to addition of the coordinates. Since the coordinates
of the identity elements are zero, the coordinates of x~ 1 must in lowest order
be (x"1)' = x'. Thus the operation of taking inverses corresponds to
taking negatives of coordinates. To see that this formula also holds in the
next order of approximation, we compute the inverse of x as a power series.
Since x o x~ 1 = e, we have
Solving for (x~ *)'. we obtain by repeated substitution,
Since cljk = clkj, the quadratic term drops out, proving our assertion.
1.7 Lie Algebras of Lie Groups 17
The commutator of two group elements x and y is the element

; it essentially determines whether x and
y commute. Expanding the commutator in a Taylor series, we have
Thus the tensor c'jt determines the predominant part of the commutator.
The local structure of a Lie group can be summarized conveniently in
terms of the tangent space to the Lie group at the identity element. Let
xl , , x", as before, be the coordinates of an element x in a neighborhood
of the identity in the Lie group. Let v(x1, , x") be the position vector of this
point with respect to the origin of the surrounding Euclidean space. Then the
vectors
form a basis for the tangent space at the identity element of the Lie group.
The coordinates x', y\ zl of the points of the Lie group close to the origin,
used in the previous expressions, can be regarded as components of vectors
in the tangent space at the identity. That is, we construct the vectors
The computations given previously can now be interpreted as operations on

vectors in this tangent space.
To the lowest order of approximation, the group operation corresponds
to vector addition in the tangent space, and taking the inverse of a group
element corresponds to taking the negative of the corresponding vector. The
operation of taking the commutator of two group elements corresponds
in the tangent space to
We can regard this as a form of vector multiplication and introduce the

notation
The properties of this vector multiplication follow from the properties of the
structure constants c'jk. It follows from the definition
definition of the structure constants
that they are antisymmetric in their lower indices,
Therefore, we have
We might ask what further properties c'jk must have. Substituting the Taylor
series for Lie group multiplication into the equation
and equating the corresponding terms, one obtains expressions relating the
c'jk to coefficients of higher order terms. By permuting the indices cyclically
and summing, one can eliminate the coefficients of the higher order terms,
obtaining the relation
Therefore, our vector multiplication also satisfies the Jacobi identity, and we
have a Lie algebra. The tangent vector space at the origin, with this multiplica-
tion, is called the Lie algebra of the Lie group. In the older literature the Lie
algebra was called the infinitesimal group. The components of the tensor c'jk
are called the structure constants of the Lie algebra with respect to the basis
e
i j ' ' ' > e -
The local structure of a Lie group, the structure in a sufficiently small
neighborhood, is completely determined by the Lie algebra. This is of great
importance in applications, for when properties of a local nature are being
studied one need only consider the Lie algebra.
1.8 VECTOR FIELDS
We can obtain another picture of the geometrical meaning of the Lie

algebra of a Lie group by using the theory of vector fields. We shall sketch this
approach here, indicating the main ideas which are involved. The general
theory of vector fields applies to any differentiable manifold, not just to
Lie groups [126]. To focus on the essential ideas, we consider here just the
simplest case of a surface in some Euclidean space which can be specified by
a single equation g(x) = 0, where g is a differentiable function. Of course, this
discussion could easily be generalized to the case of a surface specified by
several such equations.
The gradient Vg at any point of the surface can be interpreted geo-
metrically as a vector normal to the surface. It is clear that a vector U is
tangent to the surface at the point x if we have U Vg = 0 at x. The set of all
vectors tangent to a differentiable manifold at a point may be regarded as
a vector space, which we call the tangent space at that point. For a connected
differentiable manifold all tangent spaces at all points have the same dimen-
sion and this number is called the dimension of the manifold.
A vector field on a manifold is a differentiable mapping which assigns a
tangent vector U(x) to each point x on the manifold. If U(x) and V(x) are
vector fields, then it is clear that any linear combination of them will again be
1.9 Lie Theory of One-Parameter Groups 19
a vector field. Thus the set of all vector fields may be regarded as a vector
space. If we set
then
Since U Vg = V Vg = 0, it follows that W Vg = 0, and hence W(x) is

another vector field. Vector fields form a Lie algebra under this Lie product
operation, which we denote by W = [U, V].
For Lie groups we can use the action of the group on itself to associate a
certain vector field on the group to each tangent vector at the identity. A
path h(t) on a Lie group may be viewed as the trajectory of a particle moving
about on the Lie group if we regard t as the time. The velocity of the particle
at any point on the path is a tangent vector at that point. If v is a tangent
vector at the identity, we may choose a path h(t) passing through the identity
with velocity v at time t = 0. The action of an element g of the group carries
this path into a path g <> h(t) passing through the point g at time t = 0 with a
certain velocity, which is a tangent vector at g. This tangent vector at g
actually depends only on the vector v, and not on the particular choice for the
path h(t).
In this way we obtain a vector field on the Lie group, called the left-
invariant vector field determined by the tangent vector v at the identity.
These left-invariant vector fields form a vector space isomorphic to the
tangent space at the identity. Moreover, the Lie product of two left-invariant
vector fields is again a left-invariant vector field, so that these vector fields
form a subalgebra of the Lie algebra of all vector fields on the Lie group. The
Lie algebra of a Lie group may then be defined as this Lie algebra of left-
invariant vector fields on the group, this new definition being equivalent to
our old one.
1.9 LIE THEORY OF ONE-PAR A METER GROUPS
The abstract definition of a Lie group as a manifold was an outgrowth

of the way Lie groups first appeared in the theory of differential equations.
Lie groups at first arose as transformation groups and only later were
studied as abstract groups. The shift of emphasis from transformation groups
to abstract groups in the Lie theory parallels the corresponding development
which took place in the theory of finite groups. The theory of Lie groups,
originally called the theory of continuous groups, was developed by Sophus
Lie in connection with the problem of integrating systems of differential
equations. Thus, groups with analytic structure arose from the study of the
solution of differential equations, just as finite group theory arose in the
study of algebraic equations. Since Lie's work is still not widely appreciated,
we present a brief indication of his methods [63], [130].
Suppose we are presented with a first order differential equation
with analytic coefficients. There exists a one-dimensional Lie group of

transformations of the (x,j>)-plane which acts analytically and leaves the
solutions of the differential equation invariant. If we know beforehand how
this group acts on the plane, we can use this knowledge to find the solutions
of the differential equation.
An orbit is the path of a given point under the action of the group of
transformations. We say that a group acts analytically if the x and y co-
ordinates along any orbit are analytic functions of the local coordinates in
the group. We say that the solutions are invariant under the action of a group
element if it carries any given solution curve into another solution curve.
Knowing the orbits, one can introduce canonical coordinates (u, v) in the plane
such that the action of the group leaves u invariant, while v is additive in the
sense that if g t carries (u, v) to (u, v + 6^ and g2 carries (u, v) to (u, v + 62),
then gjg 2 carries (u, v) to (u, v + dl + <52).
Lie showed that in these coordinates the variables separate, so the
differential equation is reduced to quadratures. He also gave an integrating
factor for the differential equation in terms of the tangent vectors to the
orbits, from which the solution may again be obtained by quadrature.
Explicitly, if P(x, j;) and Q(x, y) are the components of the tangent vector to the
orbit through (x, y), then (PM + QN)~ l is an integrating factor. If one knows
a two-dimensional transformation group leaving the differential equation
invariant, then even the quadrature can be eliminated. In this case, the
solutions may be obtained simply by setting the ratio of two independent
integrating factors equal to an arbitrary constant.
If we associate the operator
with the group and the operator
with the differential equation, then the equation is invariant under the group
if and only if
While every first order differential equation can be shown to be invariant

under some one-dimensional Lie group, this is not the case for higher order
differential equations. When a higher order differential equation is invariant
under a one-dimensional group of transformations, one can reduce the order
of the differential equation by one upon introducing new coordinates. One
needs to find nontrivial functions u(x, y) and v(x, y, dy/dx) which are invariant
under the group. We may mention one example of this general procedure
1.10 Matrix Lie Groups 21
which frequently arises. In many applications one deals with equations which
are invariant with respect to changes of the units used for measuring the
quantities which appear in them. Then the process of finding u and v can be
carried out simply by performing a dimensional analysis of the units for these
quantities [34]. When the differential equation is formulated in terms of u
and v its order is one less than the order of the original equation. If the
dimension of the Lie group under which the differential equation remains
invariant is greater than one, the integration procedure can again be simpli-
fied by introducing canonical coordinates. These canonical coordinates and
the integration procedure depend in general on the structure of the Lie
algebra of the group of transformations and on the relative dimensions of the
group and the orbits. The dimension of an orbit may well be less than the
dimension of the group itself. For instance, the group 50(3, R) is a three-
dimensional group of transformations on three-dimensional Euclidean space,
while its nontrivial orbits are the spheres which are two-dimensional.
In addition to giving methods of reducing differential equations in-
variant under a Lie group to quadratures, Lie also showed how to find the
differential equations invariant under a given group. The problem of finding
invariant differential equations has many applications. For example, in
relativistic quantum field theory, a basic problem is that of finding all wave
equations invariant under the Lorentz group [66]. The problem simplifies
somewhat for the case of zero-mass wave equations because in that case
relativistic invariance implies in variance under the bigger group of conformal
transformations [95].
The problem of finding differential equations invariant under a given
Lie group also comes up in quite different fields. We mention here some work
by Hoffman on pattern recognition in visual perception [125]. On the basis
of physiological studies of electrical activity in the brain, he postulates that the
visual integrative process is based on a first order differential equation, with
electrical patterns in the brain embodying information about the isoclines of
the differential equation. In order for recognition of a pattern as an entity
to occur, as required by the Gestalt theory, this differential equation must be
invariant under the group of transformations which leave the pattern
recognizable. Hoffman determined the class of differential equations invariant
under a Lie group which includes translations, rotations and magnifications.
The results of this analysis are used to explain and interpret various visual
phenomena, including developmental dyslexia, the whirling spirals evoked
under flicker, the alpha rhythm and its desynchronization, and Mackay's
complementary after-images.
1.10 MATRIX LIE GROUPS

We have seen that one of the main reasons for introducing the concept
of a Lie group is to enable us to treat the classical matrix groups. If we were
only interested in these matrix Lie groups, we could even dispense with some
of the technicalities of manifold theory [73]. However, this would definitely

entail some loss of generality because there do exist Lie groups which are not
isomorphic to any matrix group [33] .
To prove that the various classical matrix groups are Lie groups, we
must show that they are analytic manifolds and that in each case the group
operation is analytic. Let us consider first the general linear groups GL(n, R)
and GL(n, C). To see that they are manifolds, we embed them in a Euclidean
space and simultaneously erect coordinate systems by noting that the n2
entries of a matrix can be used as the coordinates of a point in a Euclidean
space of dimension n2. For the group GL(n, C), we treat real and imaginary
parts of the matrix entries as separate coordinates and so use a real Euclidean
space of dimension 2n2 rather than a complex space of dimension n2. Since
the determinant is a continuous function in these coordinates, all points
sufficiently near a point representing a nonsingular matrix also represent
nonsingular matrices. Hence the general linear group is an open subset of
Euclidean space, and the Euclidean coordinates may be used as curvilinear
coordinates for the general linear group. In the general linear group, with
the chosen coordinates, the group operation is ordinary matrix multiplication
and the coordinates of the product of two matrices are polynomials in the
coordinates of the two factors. The coordinates of the inverse of a matrix
are rational functions of the coordinates of the matrix. Since polynomials and
rational functions are analytic, we see that the general linear group satisfies
the conditions required of a Lie group.
The other classical Lie groups are all subgroups of the general linear
group, and can thus also be embedded in a Euclidean space. These groups
may be regarded as hypersurfaces or intersections of hypersurfaces in
Euclidean space, so that the matrix entries are not all free to vary inde-
pendently for them. Nevertheless, in each case, appropriate curvilinear
coordinates can be found in terms of which matrix multiplication and
inversion are analytic, so that these groups are also Lie groups.
Returning to the general linear group of real or complex n x n matrices,
let us compute the structure constants by using the matrix elements as
coordinates. To have the coordinates vanish at the identity, we may express
the matrix in terms of its difference from the identity, so that for the matrix
[x'J] the new coordinates are
The product of two matrices in these coordinates is given by
Thus we have
1.10 Matrix Lie Groups 23
Therefore, the structure constants for the general linear group are given by
The tangent space has for a basis the matrices
The tangent space of the general linear group can be identified with the space
of all n x n matrices, including also those which are not invertible. If x and y
are tangent vectors (matrices) with components xu and ym", then we have
[x, y] = xy yx, where the usual matrix multiplication is meant on the
right side. We have thus determined the structure of the Lie algebra of the
general linear group of any order. The real Lie algebra of GL(n, R) is denoted
by gl(n, R), and the real Lie algebra of GL(n, C) is denoted by gl(n, C). In
general, the real Lie algebra of a Lie group may be denoted by using the
lower case letters corresponding to the letters used to denote the Lie group.
The Lie algebras of many of the classical matrix groups can be determined
by means of the exponential mapping. If each matrix in some neighborhood
of the identity in a matrix group can be expressed as exp (A) for A in some
linear subspace of the space of square matrices, then this subspace constitutes
the tangent space to the group. When endowed with the commutator
multiplication, [A, B] = AB BA, this linear subspace of square matrices
is the Lie algebra. For example, the real Lie algebra s/(n, C) of the special
linear group consists of the matrices with trace zero. The real Lie algebra
o(n, C) of the complex orthogonal group consists of all complex antisym-
metric matrices of order n, while the Lie algebra o(n, R) consists of the real
ones. Similarly, the real Lie algebra u(ri) of the unitary group consists of the
anti-Hermitian matrices. The tangent space to the intersection of two
manifolds at a point is the intersection of their tangent spaces. Hence the Lie
algebras of the matrix groups obtained by intersecting GL(n, C), GL(n,R),
SL(n, C), 0(n, C) and U(n) are simply the intersections of their Lie algebras.
For our final example, let us compute the Lie algebra so(3, R) of the
rotation group in ordinary three-dimensional Euclidean space. Our first
problem, of course, is to parametrize the matrices of the rotation group. We
can give a general formula for the matrix [atj] corresponding to an arbitrary
rotation through an angle
about the axis For any angle co ^ 0, we have
where eijk is + 1 (- 1) if (/,;, k) is an even (odd) permutation of (1, 2, 3) and 0

otherwise. There is actually no singularity at co = 0, and hence, the param-
eters CD:, co 2 , co3 can be used as local coordinates for a chart at the identity
element of the group S0(3, R). The tangent space at the identity has as a basis
the three matrices ek having (y)-entry
Thus,
and
Then an element of the Lie algebra so(3, R) is an antisymmetric matrix
The Lie product is given by
If we identify the vectors e-t with the usual unit vectors of vector analysis, we
recognize this to be the ordinary vector cross product. Therefore, the Lie
algebra so(3, R) of the rotation group is isomorphic to the three-dimensional
real Euclidean space with the vector cross product as the Lie multiplication.
Of course, we could equally well have arrived at the cross product algebra in
this case by using the exponential mapping.
1.11 POISSON BRACKETS
Lie algebras play a fundamental role in the study of conservation laws

in both classical and quantum mechanics. We briefly sketch here the general
framework of the classical theory of conservation laws, using Poisson
brackets.
Let us recall how Poisson brackets are used in classical mechanics to
rewrite Newton's equations of motion in a canonical form. Newton's laws
give differential equations for the trajectory of a mass point moving in three-
dimensional Euclidean space or along some curve or surface in Euclidean
space. We can also deal with a system of N mass points moving in three-
dimensional Euclidean space by regarding them as a single point in a
1.11 Poisson Brackets 25
3N-dimensional space. Many problems in classical mechanics are best

treated in terms of curvilinear or generalized coordinates q1, , q" instead
of in rectangular Cartesian coordinates. In this case, Newton's laws for a
system of particles may be considered as equations of motion for a single
point in a configuration space, which we assume to be a differentiate mani-
fold. As time varies, the point representing the system moves along a path in
configuration space called the trajectory of the system. Newton's equations
involve both velocity and acceleration, and are thus second order differential
equations, but they can be replaced by a set of first order differential equations
if we introduce the velocities q' dql/dt as additional variables. The position
and velocity of the system can then be specified by a set of 2n coordinates
(q1, , <f, q1, , 4") fr a point on another manifold called the tangent
bundle of the configuration space [1]. In these coordinates, Newton's equations
may be rewritten in a certain canonical fashion by using the Lagrangian
formalism.
The Lagrangian L^q1, - , q", q1, , 4") of a system of particles may
be regarded as a smooth function defined on the tangent bundle manifold.
The trajectories of a mechanical system are paths on this manifold satisfying
Lagrange's equations,
Lagrange's equations have the same form in any coordinate system because
they are the equations for the extremal curves which minimize a certain
integral.
To obtain an even more symmetric formulation for classical mechanics,
we must pass from this Lagrangian description to the Hamiltonian descrip-
tion via the Legendre transformation. We introduce the generalized momenta
PJ associated with the corresponding coordinates qj by
where j = 1, , n. If such an elimination is not possible, it is said that the

system has constraints [77], [78]. The space with coordinates (q1, , q",
Pi > '" > Pn) is called the phase space. The Hamiltonian H is a function defined
in phase space by
A given physical system is described by a trajectory in phase space determined

by Hamilton's equations of motion, together with suitable initial conditions.
Hamilton's equations
are not valid for an arbitrary system of coordinates in phase space. They
hold only for certain sets of canonical coordinates which split into two sets, a
set of position coordinates and the corresponding momentum coordinates.
By a dynamical variable, we mean any infinitely differentiable function
of the coordinates, momenta and time. The dynamical variables form an
infinite-dimensional real Lie algebra with the vector multiplication
The Lie product [A, B] of two dynamical variables A and B is a dynamical

variable called the Poisson bracket of A and B. By direct computation one can
show that the Poisson bracket satisfies the three axioms for a Lie product:
bilinearity, antisymmetry and the Jacobi identity [138]. The Lie algebra of all
dynamical variables contains no physical information about the system
beyond the number of degrees of freedom since all systems with the same
number of coordinates and momenta have isomorphic Lie algebras.
However, the subalgebra of conserved quantities, describing the physical
symmetries of the system, is usually of considerable interest. A dynamical
variable A is conserved itdA/dt 0 is valid identically along every trajectory.
In general, the rate of change of a dynamical variable A while the system
moves along its trajectory is given by
In particular, if the Hamiltonian is not explicitly a function of time, then it is

conserved, and if there are no constraints, this is simply the law of energy
conservation. For any two dynamical variables A, B the preceding equation
can be used together with the Jacobi identity to show that
It follows that if A and B are conserved, then the Poisson bracket of A and B
is conserved. Also any real linear combination of conserved quantities is
conserved, and hence the set of all conserved dynamical variables forms a
subalgebra of the Lie algebra of all dynamical variables.
1.12 QUANTUM SYMMETRIES

The discussion of symmetries and conservation laws in quantum
mechanics differs considerably from the classical theory. In quantum
mechanics, the states of a system are not described by points in a phase space,
but are instead described in terms of Hilbert spaces. A Hilbert space is a
vector space (usually infinite-dimensional) over the complex number field
equipped with an inner product and having the property that every Cauchy
sequence of vectors converges. The inner product of two vectors *F and O is
1.12 Quantum Symmetries 27
denoted by (HP, O). By a linear operator in a Hilbert space we mean a linear

transformation whose domain and range are subspaces of the Hilbert space.
If a linear operator A has a dense domain, we can define its adjoint A* as
follows [80] : The domain of A* is the set of all vectors * for which (*F, XO) is
a continuous linear function of 3> on the domain of A. The linear operator A f
is then defined on this domain by the requirement that
for all $ and *F in the domains of A and A\ respectively.

An extension of a linear operator is a linear operator defined on an
extended domain and agreeing with the original operator on the original
domain. A linear operator H is Hermitian if Hf is an extension of H, and we
say that H is self-adjoint if H = H\ A unitary operator U in a Hilbert space is
a linear operator defined on the whole Hilbert space satisfying
Thus, a unitary operator has the property that
for all vectors HP and <i> in the Hilbert space. We also have to deal with anti-
unitary operators, which are not linear operators at all, but are instead
antilinear. An antilinear operator A in Hilbert space is a mapping which
preserves vector addition, but satisfies
for all complex numbers c and vectors 4/. Here c* denotes the complex
conjugate of the number c. An antiunitary operator A is an invertible anti-
linear operator which satisfies
for all vectors O and *F in the Hilbert space.

We now recall some of the basic ideas and assumptions of quantum
mechanics. A physical system may be regarded as a reproducible experimental
apparatus, and a state of a system is the result of a preparation of the system,
that is, the previous history of the system. Systems prepared in identical
fashions will typically yield different values for measured quantities, and
therefore all measurements on systems are to be understood in a statistical
sense only. Two methods of preparing the system are considered to lead to
the same state if they lead to identical probability distributions for all possible
measurements. Thus the state of a system is not really a specific history, but
rather an equivalence class of such histories. In quantum mechanics one
usually distinguishes between pure states and mixtures. Since the description
of mixtures can be reduced to that of the pure states, it is sufficient to deal
with the latter. It is assumed that each pure state of a system corresponds to
at least one unit vector in a Hilbert space.
If, conversely, every unit vector also corresponds to some physical state,
we say that there are no superselection rules [227]. The relation between pure
states and unit vectors is not one-to-one since it is assumed that unit vectors
differing only by a phase factor will correspond to the same state. In the
absence of superselection rules, we can describe the relation between Hilbert
space and pure states as a mapping of the unit sphere in Hilbert space onto
the set of pure states. If O and T are unit vectors, then |(<I>, )|2 is interpreted
as the probability that a system prepared in the pure state corresponding to
<X> will be found upon measurement to be in the pure state corresponding to
*F. Note that states which are indistinguishable by these transition probabili-
ties are considered to be identical states.
A physical symmetry is an onto mapping of pure states to pure states
which preserves all transition probabilities. The assumption about in-
distinguishability implies that physical symmetries must also be one-to-one
as mappings and hence have inverses. The products and inverses of physical
symmetries are again symmetries, and hence the physical symmetries form a
group. Wigner discovered that physical symmetries correspond to unitary or
antiunitary operators in Hilbert space [19], [249]. This correspondence is
again unique only to within phase factors. The group generated by these
unitary and antiunitary operators will therefore in general be an extension
of the physical symmetry group. Technically, the physical symmetry group
will be isomorphic to the quotient group of this group of operators modulo
the group of phase factors [211].
Although the descriptions of classical and quantum symmetries appear
to be very different, the classical description must of course be a limiting case
of the quantum description. There actually is a close relation between the
classical and quantum descriptions of dynamics, which historically played an
important role in developing quantum mechanics via the correspondence
principle. The correspondence principle was used to guess the quantum
analogues of classical dynamics in the early days of quantum mechanics.
Any such method of assigning quantum analogues to classical dynamical
variables may be called a quantization procedure. Starting with the methods
proposed in 1926-1927 by P. A. M. Dirac, J. von Neumann and H. Weyl, a
large variety of quantization procedures have been studied over the years
[210]. One of the most useful of these quantization methods, due to Dirac,
makes use of Lie algebraic ideas. The Dirac correspondence principle relates
the Poisson brackets of classical mechanics to quantum mechanical com-
mutators [75], [76], [79]. In this theory the classical Lie algebra of dynamical
variables is related to a Lie algebra of operators in Hilbert space in quantum
mechanics. By using this correspondence principle, we can often discuss
symmetries of analogous systems in a parallel fashion. For example, the
Kepler problem of planetary motion and the hydrogen atom are analogous
systems and so have the same symmetries (the four-dimensional orthogonal
group), which we discuss further later on.
1.12 Quantum Symmetries 29
The quantum mechanical analogue of the classical dynamical variables

is a set of self-adjoint operators in Hilbert space called the set of observables.
For the constant dynamical variables the corresponding operators are
constant real multiples of the identity operator / in Hilbert space. A measure-
ment of an observable A will yield a definite value, say a, only in states de-
scribed by eigenvectors of A corresponding to the eigenvalue a. We further
assume that the observables form a real Lie algebra if the Lie product of
observables A and B is defined as a purely imaginary multiple of the com-
mutator AB BA. The possibility of making such an assumption depends,
of course, on the existence of a fixed common dense domain in the Hilbert
space for all of the observables [96] , [97] .
The correspondence principle requires that we define the Lie product of
observables A and B to be
where h is the quantum of action (Planck's constant divided by 2n). The

relation between classical and quantum mechanics found by Dirac may be
formally described as a homomorphism from part of the classical Lie algebra
of dynamical variables to part of the quantum Lie algebra of observables.
To illustrate this, let us consider the description of a nonrelativistic point
particle. If xf and p, are the ordinary Cartesian coordinates and momentum
components of the particle, then the classical Poisson brackets are
and
The corresponding operator equations in quantum mechanics must be
and
where X( is the operator corresponding to x,, and so forth. If we take X, to

be the linear operator corresponding to multiplication of any function
ij/(\) by jc,-, and P{ to be the differential operator ihd/dx^ then a purely
formal computation indicates that these commutation relations are fulfilled.
In the Schrodinger formulation of nonrelativistic quantum mechanics, these
operators are interpreted as Hermitian operators on an appropriate Hilbert
space of equivalence classes of square-integrable functions.
The quantization prescription given above is unfortunately not com-
plete. The requirement that the quantization map be a Lie algebra homo-
morphism does not define it uniquely on the whole set of dynamical variables.
To obtain a complete quantization prescription, the ambiguities which remain
have to be removed by additional assumptions [108], [250], [251]. In the

last analysis, of course, there is no reason to expect that any quantization
procedure will be entirely satisfactory for physics. Strictly speaking, the
correspondence principle only requires that the classical limit of quantum
mechanics be derivable in some precise way, but not necessarily the other
way around. Nevertheless, these quantization procedures may be useful
to construct mathematical models by which one can test the consistency of
some systems of axioms proposed for quantum mechanics. For example,
there has been some interest in using quantization procedures to exhibit a
nontrivial example of a relativistic quantum field theory [137], [214], [223].
1.13 HARMONIC OSCILLATORS
Harmonic oscillators are important in both classical and quantum

mechanics because conservative stable linear systems have positive definite
quadratic Hamiltonians and are therefore formally equivalent to collections
of harmonic oscillators. For example, electromagnetic radiation is described
by the linear Maxwell equations and may be treated as a collection of
harmonic oscillators. Again, a dilute gas consisting of a large number of
particles may be treated equivalently as a system of sound waves in a con-
tinuous medium. Harmonic oscillators are also the basis for many models in
molecular, atomic and nuclear physics. Our interest here lies in the fact that
the unitary groups arise naturally in the discussion of systems of identical
harmonic oscillators [15], [20], [173]. This symmetry is broken if there is any
interaction among the oscillators, but cannot be ignored if we use pertur-
bation theory to describe the interaction.
It is easy to exhibit the unitary symmetry of a system of harmonic
oscillators directly. We adopt the quantum mechanical point of view because
it is simpler, but the same type of discussion can be given in classical mechanics.
The Hamiltonian for a system of n identical noninteracting harmonic
oscillators is
In discussing harmonic oscillators, it is natural and convenient to choose

units in which m = k = h = 1. We define creation and annihilation operators
by
in terms of which the Hamiltonian may be written as

1.14 Lie Subgroups and Homomorphisms 31
The creation and annihilation operators satisfy the commutation relations
with each other, and
with the Hamiltonian. These latter commutation relations imply that if ij/
is an eigenvector of H with eigenvalue E, then ax\}/ is either zero or an eigen-
vector with eigenvalue E 1, while a* if/ is an eigenvector with eigenvalue
E + 1. Thus the harmonic oscillator has an infinite series of equally spaced
levels.
Both the Hamiltonian and the commutation relations are invariant
under a unitary transformation,
so U(n) is a symmetry group of the n-dimensional harmonic oscillator.

The corresponding Lie algebra u(n) is spanned by the elements a*a.p, which
commute with the Hamiltonian operator H.
In classical mechanics, the Hamiltonian
appears at first sight to be symmetric under the orthogonal group 0(2n, R).
However, one must remember that Hamilton's equations hold only for
certain canonical systems of coordinates in phase space. In the case of linear
transformations of the coordinates (ql, , q", pj , , pj in phase space,
this requirement limits us to symplectic linear transformations [65] . In view
of the isomorphism between U(n) and Sp(n, R) fl 0(2n, R), we see that the
symmetry group of a system of n identical classical harmonic oscillators is
again the unitary group [120].
1.14 LIE SUBGROUPS AND ANALYTIC HOMOMORPHISMS

For many of the problems of mathematics and physics in which Lie
groups arise, the most interesting information can be derived by looking at
their Lie algebras. Therefore, we now return to the study of the relation
between Lie groups and Lie algebras, to prepare for a more extensive study
of Lie algebras.
In general, a subgroup of a Lie group need not be a Lie group. In order
for a subgroup to be a Lie group, it must possess its own atlas of local co-
ordinate systems, in terms of which the group operations are analytic. A Lie
subgroup of a Lie group is a subgroup which is itself a Lie group and for which
the local coordinates in the subgroup are analytic functions of those in the
group. The Lie algebra of a Lie subgroup of a Lie group may be naturally
identified with a subalgebra of the Lie algebra of the original group. Con-
versely, every subalgebra of the Lie algebra of a Lie group may be identified
with the Lie algebra of a connected Lie subgroup. Every closed subgroup of a
Lie group is a Lie subgroup. However, a Lie subgroup of a Lie group need
not be a closed subgroup. From the theory of topological groups it is known
that every open subgroup is closed, and hence open subgroups of Lie groups
are also Lie subgroups [64].
Similarly, a homomorphism between Lie groups in general need not be
an analytic mapping. A homomorphism between Lie groups is called an
analytic homomorphism if the coordinates of the image of a point are analytic
functions of the coordinates of the point. Analytic homomorphisms of Lie
groups induce homomorphisms of the corresponding Lie algebras. The
homomorphism induced by a composition of analytic homomorphisms
is the composition of their induced homomorphisms. This relation between
homomorphisms of Lie groups and Lie algebras can be used to translate
many results about Lie groups into related results about their corresponding
Lie algebras. For example, the inclusion mapping for a Lie subgroup is an
analytic homomorphism, and hence induces the corresponding embedding
for their Lie algebras noted above. It is a remarkable fact that any con-
tinuous homomorphism of Lie groups is also analytic [124].
If H is a closed normal subgroup of a Lie group G, then we may regard
the quotient group G/H as a Lie group in a natural way, and the natural
projection is analytic since it is known to be continuous. Finally, as one might
expect, a form of the fundamental theorem of homomorphisms holds for Lie
groups, and it is related to the corresponding theorem for their Lie algebras.
The kernel of an analytic homomorphism of a Lie group is a closed normal
subgroup. The kernel of the induced Lie algebra homomorphism is the Lie
algebra of the kernel of the group homomorphism.
1.15 CONNECTED LIE GROUPS
To a large extent, even the global structure of a Lie group is determined

by its local structure, that is, by what happens in an arbitrarily small neigh-
borhood of the identity. This is because by multiplying together many
elements very near to the identity element, we can obtain elements further
away. Also, any neighborhood of the identity can be transported along any
arc in arbitrarily small steps, much as one does analytic continuation. One
may therefore ask whether the whole Lie group is determined by its Lie
algebra, the answer being yes, provided that the Lie group is simply-connected
[54], [192]. (The term simply-connected and several other terms used in the
following discussion are defined below.) Each real Lie algebra is isomorphic
to the Lie algebra of a simply-connected Lie group, which is uniquely deter-
mined to within an isomorphism. Hence, a simply-connected Lie group is
1.15 Connected Lie Groups 33
completely determined by its Lie algebra to within an isomorphism. The

proof of this theorem is based on the existence of solutions to certain systems
of partial differential equations. The global structure theory for an arbitrary
Lie group almost reduces to the study of two discrete Lie groups and one
simply-connected Lie group. This reduction proceeds in two stages, the first
stage involving the extraction of the identity component, and the second
involving passing to the universal covering group. Many of the ideas here
are based on the more general theory of topological groups, but for Lie
groups some simplifications occur. Lie groups, being manifolds, are locally
homeomorphic to open regions of Euclidean space, and therefore have very
nice local properties. In particular, every Lie group is a locally compact
normal Hausdorff space.
A topological space is connected if it has no proper subsets which are
both open and closed. An arc y in a topological space is a continuous mapping
from the closed interval [0,1] into the space, and a loop is an arc for which
y(0) = y(l). A topological space is arcwise connected if any two points of the
space can be joined by an arc in the space. A manifold is connected if and only
if it is arcwise connected, and hence the same is true for Lie groups. A Lie
group is simply-connected if it is connected and every loop in the group can be
continuously shrunk to a point. The real line, regarded as a group under
addition, and the unit circle in the complex plane, regarded as a group under
multiplication, are both examples of connected Lie groups.
At the other end of the spectrum of Lie groups are the discrete groups.
A topological space is discrete if every subset is open. A discrete topological
group is totally disconnected in the sense that no two distinct points can be
joined by an arc. Any discrete topological group may be regarded as a Lie
group by taking the charts to consist of single points and assigning zero as the
coordinate of any point. The Lie algebra of a discrete Lie group consists only
of the zero vector, and conversely, if the Lie algebra of a Lie group is zero,
then the group is discrete.
A component of a Lie group is a maximal connected subset, that is, a
subset consisting of all elements which can be joined by arcs to some given
element. The component containing the identity element of a Lie group is a
closed normal subgroup, and the components of the Lie group are precisely
the cosets of this subgroup. Moreover, the identity component G0 of a Lie
group G is an open subgroup, so that the quotient group G/G0 is not only
totally disconnected, but even discrete.
For example, the real general linear group is not connected since there
is a surface separating the elements with positive determinants from those
with negative determinants. We may picture the group GL(n, R) as filling
a real Cartesian space of dimension 2, except for the surface consisting of the
points corresponding to the singular matrices, which divides the space into
two connected components. The identity component consists of the elements
with positive determinant, and the discrete quotient group in this case is the
cyclic group of order two.
The study of a Lie group G does not quite reduce to studying the con-
nected identity component G0 and the discrete quotient group G/G0. One
also needs to know the action of certain inner automorphisms restricted to
the identity component to obtain the complete structure of a Lie group.
Frequently, for Lie groups of practical importance, the structure of the
quotient group is extremely simple, and knowledge of it together with
knowledge of the structure of the identity component is all that is needed to
completely determine the group structure.
For the second stage of the reduction of the global structure of a Lie
group, our analysis is directed toward the structure of the identity component
G0, which is a connected Lie group. The Lie algebra of the original Lie group
is isomorphic to the Lie algebra of the identity component. This Lie algebra
is also isomorphic to the Lie algebra of a simply-connected Lie group, which
is called the universal covering group of the connected Lie group G 0 . The Lie
group G0 is the homomorphic image of the simply-connected universal
covering group determined by the Lie algebra. The kernel of this homo-
morphism is a discrete normal subgroup of the universal covering group
called the fundamental group or Poincare group [36], [200], [205]. Each point
of this discrete fundamental group corresponds to a set of loops in the
connected group G 0 . Two such loops correspond to the same point of the
kernel if and only if one can be continuously transformed into the other.
1.16 ABELIAN LIE GROUPS
We may illustrate some of the results of the global structure theory for
Abelian Lie groups. A Lie group is Abelian if
for all elements a and b of the group. The simplest Abelian Lie groups are the
vector groups. A vector group is a real vector space R", regarded as an
Abelian Lie group under the operation of vector addition, using any
rectilinear coordinate system to provide a chart. Vector groups are simply-
connected, and conversely, every simply-connected Abelian Lie group is
isomorphic to a vector group. An Abelian Lie algebra is a Lie algebra whose
structure constants are all equal to zero, so that
for all elements x and y of the Lie algebra. Note that any real vector space may
be made into an Abelian Lie algebra by introducing this trivial multiplication.
We can now use the theory of the universal covering group to classify
all the connected Abelian Lie groups. The Lie algebra of any Abelian Lie
group is an Abelian Lie algebra, and we may regard the Lie algebra itself as
a vector group. Obviously, a vector group is its own tangent space, and so
with the trivial multiplication, is its own Lie algebra. Hence, every connected
1.17 Low-Dimensional Lie Groups 35
Abelian Lie group has a vector group as its universal covering group. Every
connected Abelian Lie group is then a homomorphic image of a vector group,
the kernel being a discrete subgroup. The discrete subgroups of a vector
group are crystallographic lattices of vectors of the form
where the n, run over all integers and the e, are fixed linearly independent
vectors.
Thus every connected Abelian Lie group is a direct product of lines
and circles. The one-dimensional simply-connected Abelian Lie groups are
isomorphic to the additive group of the real line. Any discrete subgroup of the
real line consists of equally spaced points, all the integer multiples of a fixed
real number. Identifying points of the line modulo this discrete subgroup is
equivalent to winding the line about a circle of circumference equal to the
spacing between the points of the subgroup. If k is the least positive member
of the discrete subgroup, then
maps the real line homomorphically onto the circle group, and has the
discrete subgroup as kernel. Thus, the circle group has the real line group as
its universal covering group, and the discrete fundamental group of the circle
group is isomorphic to the additive group of integers. In short, the line and the
circle are the only connected one-dimensional Abelian Lie groups. The only
two-dimensional connected Abelian Lie groups are the plane, the cylinder
and the torus. Note that the plane group may also be regarded as the additive
group of all complex numbers, and the cylinder group as the multiplicative
group of the nonzero complex numbers. For higher dimensions, the results
are similar, there being exactly n + 1 different connected w-dimensional
Abelian Lie groups.
1.17 LOW-DIMENSIONAL LIE GROUPS
Many Lie groups may be realized as matrix Lie groups. We recall that
analytic subgroups of a Lie group correspond to subalgebras of its Lie
algebra, and the Lie algebra may be taken to be the tangent space at the
identity. The tangent space of a matrix Lie group is some subspace of the
tangent space of the general linear group, the latter being identified with the
space of all square matrices of a given order. The Lie product in the tangent
space of a matrix Lie group is always the commutator, just as it is for the
general linear group.
To illustrate this, let us consider the real affine group consisting of all real
matrices of the form
We may think of this real affine group as the identity component of the group
of inhomogeneous linear transformations on the real line [11], [12]. Since the
underlying manifold of the real affine group is a plane with coordinates (a, b),
the group is a simply-connected Lie group. A basis for the tangent space
consists of the matrices
The Lie product is simply
In fact, any non-Abelian two-dimensional real Lie algebra has this structure
for an appropriate choice of basis. Since this affine Lie algebra is the only
non-Abelian two-dimensional one, any connected two-dimensional non-
Abelian Lie group is isomorphic to the real affine group modulo a discrete
normal subgroup. By writing out the form of a conjugate class, it is easily
seen that no proper normal subgroup can be discrete. The real affine group
is therefore the only connected non-Abelian two-dimensional Lie group.
Thus, the only possible connected two-dimensional Lie groups are the plane,
the cylinder and the torus. The real Cartesian plane R2 admits two group
structures making it a Lie group, one Abelian and one non-Abelian. The
torus and the cylinder admit only an Abelian Lie group structure.
For higher dimensions, finding all connected Lie groups becomes a
much more complicated problem. A part of the solution, finding all the
possible real Lie algebras of a given dimension, has been studied [174] , [175] ,
[176], [177]. We shall say more about this part of the problem later on.
1.18 THE COVERING GROUP OF THE ROTATION GROUP

Since S0(3, R) is connected but not simply-connected, it is the homo-
morphic image of a simply-connected Lie group having the same Lie algebra.
The simply-connected Lie group associated with the vector cross product
Lie algebra is the special unitary group SU(2). The homomorphism of 5(7(2)
onto S0(3, U) is of fundamental importance in the Pauli theory of electron
spin.- The Pauli matrices are a set of 2 x 2 matrices,
ff
The group SU(2) then consists of all 2 x 2 matrices of determinant one

which can be written in the form
where a, b, c and d are real numbers, and 1 is the identiy matrix. Since the
determinant of such a matrix is a2 + b2 + c2 + d2, we see that the group
1.18 The Covering Group of the Rotation Group 37
manifold of SU(2) can be identified with the unit sphere in a real four-
dimensional Euclidean space with coordinates (a,b,c,d). Thus the special
unitary group SU(2) is a simply-connected three-dimensional Lie group [54].
For any Q in 5 7(2), we define a 3 x 3 matrix [atj] by
where Q* is the Hermitian conjugate of Q. The matrix [aj is a rotation

matrix, and every rotation matrix can be obtained from a matrix Q in SU(2)
using this formula. Moreover, two distinct members Q and Q' of the group
SU(2) yield the same rotation matrix if and only if Q' - -Q. For any pair
of traceless 2 x 2 matrices X, Y we have the identity
This identity can be used to verify that the mapping of SU(2) onto 50(3, R) is
a group homomorphism. By the fundamental homomorphism theorem,
the rotation group 50(3, R) is isomorphic to the quotient group of SU(2)
modulo its center, the discrete subgroup consisting of the identity and its
negative. Incidentally, since the group 5(7(2) is homeomorphic with the
unit sphere in 4-space, it is clearly compact, and the rotation group 50(3, R),
being a continuous image of 5L^(2), is also compact.
The relationship between the groups 50(3, R) and 517(2) can also be
made explicit in the following alternative manner. The elements of 50(3, R)
correspond in a one-to-one fashion with the rotations of a sphere about its
center. The points of the sphere other than the "north pole" may be identified
with the complex plane by the usual projection from the "north pole" onto a
plane tangent to the sphere at the "south pole." Under this projection, each
rotation of the sphere induces a corresponding mapping of the complex
plane onto itself, a complex function of a complex variable. The complex
functions found in this way are certain linear fractional transformations of
the form
where a, /?, y and d are complex numbers and a<5 /?y ^ 0. The composition
of two such linear fractional transformations is again a linear fractional
transformation. Moreover, the condition a<5 /?y ^ 0 ensures that the
transformation has an inverse, which is again a linear fractional transforma-
tion.
The linear fractional transformations form a group, and the rotation
group is isomorphic to a subgroup of this group. Obviously /(z) is unchanged
if all the coefficients a, /?, y, 6 are multiplied by a common factor, so by
introducing an appropriate factor we can always make ad yf$ 1. Since
only the square of a common factor enters into the expression a<5 y/?, its
negative will serve equally well to satisfy the condition ad - yp = 1. To

each such /(z), therefore, there correspond two matrices of coefficients,
and its negative, both with determinant one. The linear fractional transforma-
tion corresponding to a product of two such matrices is the composition of
the linear fractional transformations corresponding to the matrices being
multiplied. Thus we have a homomorphism of the group SL(2, C) onto the
group of linear fractional transformations, and the kernel of this homo-
morphism consists of the identity matrix and its negative. Not every linear
fractional transformation, however, corresponds to a rotation of a sphere.
In fact, the group of linear fractional transformations is isomorphic to the
identity component of the Lorentz group. A linear fractional transformation
corresponds to a rotation of the sphere if and only if it corresponds to a
unitary matrix. That is, its matrix must have the form
where a, b, c and d are real numbers. The condition a<5 /?y = 1 is precisely
the condition that the vector (a, b, c, d) lie on the unit sphere in 4-space.
Thus again we obtain the homomorphism of SU(2) onto S0(3, R) as before.
Diagram 1 summarizes the interrelations between the various groups
discussed here.
Here we have denoted the linear fractional transformation group, or pro-

jective special linear group, by PSL(2, C).
1.19 TENSOR PRODUCT OF VECTOR SPACES
The study of the local structure of Lie groups, based on the theory of
Lie algebras, makes heavy use of linear algebra and vector space concepts.
For the complexification process, as well as in connection with the theory
of bilinear forms and elsewhere, it is convenient to use a basic concept of
vector space theory, the tensor product. We may recall that in the
discussion of the geometric concept of orthogonality and in the definition of
1.19 Tensor Product of Vector Spaces 39
associative and Lie algebras, we already had to deal several times with bilinear
functions.
There are actually several different, but equivalent, ways to define the
tensor product of two vector spaces V1 and V2 . The best way is to give a
universal definition, defining the tensor product by means of its properties.
Another way, which assures the existence of the object being defined, is to
give a completely constructive definition, and this is what we shall do first.
We temporarily use the nonstandard but rather suggestive notation v1 x v2
for the ordered pair (i^ , v2) having for its first element a vector vl in V^ and
for its second element a vector v2 in V2 . The set of all such ordered pairs
can be regarded as the basis of an infinite-dimensional vector space consisting
of all their finite formal linear combinations. A typical vector in this space
is a vector
Consider the set of all vectors in this space of the forms
and
The quotient space of the whole infinite-dimensional space with respect to

the infinite-dimensional subspace spanned by these vectors is called the
tensor product vector space Vt V2 . In this construction, the elements of the
vector space V1 V2, which we may naturally call tensors, are of course
cosets of vectors in the original space with respect to the subspace. In par-
ticular, the coset determined by the ordered pair vl x v2 is a tensor denoted by
v1 <g) v2 which we call the tensor product of the vectors vl and v2.
It follows from the construction that
for any scalar a, and that
If {vlt} is a basis for Vl and {v2j} a basis for V2, then the set of tensors
{î, v2j} is a basis for V1 (g> V2. Thus the dimension of the tensor product
Vl V2 is the product of the dimensions of V^ and V2 , and any tensor can be
written as
The array of coefficients a0- is uniquely determined by the tensor being

represented, and it changes appropriately when the bases in Kt and V2 are
changed [2 15].
Finally we may note that the construction given above is not the only
one possible. Another way to construct the tensor product V^ (g) V2 can be
given in terms of the dual space of the vector space of all bilinear forms on
the Cartesian product V x V2. For an ordered pair (vl,v2) in Vl x V2
we let vt (g) v2 be the element of this dual space for which
for all bilinear forms ft on V1 x V2 . The tensor product Vl V2 may then be

defined as the subspace spanned by the set of all such elements v (g) v2. In
the case of finite-dimensional vector spaces, the subspace V V2 is in fact
the whole dual space of the space of bilinear forms, while in the infinite-
dimensional case it is a proper subspace of this dual space [114], [136].
If K! and V2 are vector spaces, then a function ft : Vl x V2-^W mapping
K! x V2 into W is called a bilinear mapping if it satisfies the following pro-
perties for any scalars a and b and any vectors x, y, u in Vl and u, w, z in V2 :
Bilinear forms are of course obtained as a special case of bilinear

mappings if we replace the vector space Wby the base field. A useful property
of the tensor product is that any bilinear mapping W determin
a unique linear mapping a > W such
for all vl'mVl and all v2 in V2 . The process of replacing the bilinear mapping ft
by the linear mapping a is called lifting, and we say that ft is lifted to a.
Some special cases of the process of lifting bilinear mappings to linear
mappings should be mentioned. Any pair of linear mappings between vector
space determines a linear mapping between the corresponding tensor
product spaces. If a t : V -> V\ and a2 : V2 - V2 are linear mappings, then we
first obtain a bilinear mapping from Vt x V2 to V\ (g) V2 which takes the
ordered pair (i^, v2) into the tensor a^) (g) a 2 (u 2 )- This bilinear mapping
can then be lifted to a linear mapping denoted by a t a2 called the tensor
product of the linear mappings tx.1 and a 2 . The linear mapping o^ a 2 :
Vl (g) V2 -* Ki (g) F2 thus obtained has the fundamental property that
If F is a vector space over the scalar field F, then the mapping which takes
(a, v) into au is a bilinear mapping F x V -* V, which can be lifted to a linear
mapping F (g) V -> V. This linear map is an isomorphism which may be used
1.19 Tensor Product of Vector Spaces 41
to identify F V with V, and similarly we may also identify V F with V.

A linear form on a vector space V over a scalar field F is a linear mapping
from V into F. Every finite-dimensional vector space V determines another
vector space of the same dimension, the dual space V* consisting of all linear
forms on V. The function which takes (, v) into (v) is a bilinear mapping
from V* x Kinto F, and consequently can be lifted to a linear mapping from
V* Finto F. The elements of V* V are often called mixed tensors of rank
two, and this linear mapping from V* (g> V into F is then called a contraction
of such tensors.
If e , , en is a basis for V, then the dual basis el, - , e" for V* consists
of the linear forms on V satisfying
A mixed tensor in V* (g) V may then be written as
The subscript and superscript on a/ are called covariant and contravariant

indices, respectively. Contraction of this tensor yields the scalar ,-a/.
Thus contraction in this notation corresponds to setting the upper index of
the coefficient array equal to the lower index and summing on it. It is clear
that contraction very much resembles the operation of taking a trace. Indeed,
they become the same when the spaces V* Fand linF (V) are identified in a
certain natural way.
The process of complexification for a real Lie algebra can be explained
as follows. We may recall that a real or complex vector space with a bilinear
vector multiplication satisfying the properties
and the Jacobi identity
is called a Lie algebra. If e ^ , , en is a basis for a real Lie algebra L, then the
Lie product is determined by the structure constants cy* defined by
Since the definition of a Lie algebra amounts to imposing restrictions on the

structure constants, any real Lie algebra L with structure constants c^ can
be extended to a complex Lie algebra having the same structure constants.
More formally, if L is a real Lie algebra, then we can form the tensor product
space C L, where the complex number field C is here regarded as a two-
dimensional real vector space. The vector space C L may be regarded as a
complex vector space since we can define
for all A, fj. e C and all x e L. Finally, the complex vector space C L may
be regarded as a complex Lie algebra if we set
this complex Lie algebra being called the complexification of L. Conversely,

a complex Lie algebra L may be regarded as a real Lie algebra LR, since the
multiplication of Lie elements by real numbers may be defined simply as the
obvious restriction of the multiplication by complex numbers. This Lie
algebra LR is called the real restriction of the complex Lie algebra L. Thus we
can pass fro;n real Lie algebras to complex Lie algebras, and back again.
1.20 DIRECT SUMS OF VECTOR SPACES
The direct sum, like the tensor product, is a fundamental vector space
operation which finds many applications in the theory of Lie algebras and
their representations. The direct sum operation in vector space theory is
useful both as an analytical tool and as a constructive procedure. Corre-
sponding to these two modes of usage, there are actually two slightly different
definitions of the direct sum, known as the internal and the external direct
sum. In practice there is little danger in being a bit careless on this point since
these two variants are to a large extent equivalent, and the distinction between
them can usually be understood from context. If Vl and V2 are vector spaces
over a field F, their external direct sum consists of all the ordered pairs
(vl, v2), where v: is in Vl and v2 is in V2. The external direct sum, denoted by
Vl + V2, may be regarded as a vector space if vector addition and multiplica-
tion of vectors by scalars are defined componentwise. A familiar example of
the use of the external direct sum is in the construction of the n-dimensional
vector space
over a field F. Another example of the use of the external direct sum is in the
construction of the tensor algebra, to be discussed later.
The concept of internal direct sum is used when we are talking about
the lattice of all subspaces of a given vector space. The set of all subspaces of
a vector space is closed under the operation of intersection, but not under the
operation of union. The sum S: + S2 of subspaces Si and S2 of a vector space
is the set of all elements of the form x + y, where x is in S x , and y is in S 2 .
Equivalently, the sum of two subspaces may be described as the subspace
spanned by their union. The set of all subspaces of a vector space is said to be
a lattice under the two operations D and +. This means that these operations
satisfy certain axioms somewhat reminiscent of Boolean algebra, but not
quite as strong [112], [134]. In particular, the distributive laws between fl
and + do not hold, and there is no analogue of the de Morgan laws of com-
plementation. More formally, a lattice is a partially ordered set in which
1.21 The Lattice of Ideals of a Lie Algebra 43
every pair of elements has a least upper bound and a greatest lower bound.
In the case of the lattice of subspaces of a vector space, the partial ordering is
just the inclusion relation, while S t fl S2 is the greatest lower bound of Sl
and S2 and Sl + 52 is the least upper bound of S t and S2.
The role of complementation in set theory is replaced in vector space
theory by the concept of supplement, which is defined in terms of the internal
direct sum as follows. The sum of two subspaces is said to be an internal
direct sum when the intersection of the two subspaces consists of the zero
vector. If Vis the internal direct sum of subspaces S and T, written V S T,
then every element in V can be written in a unique fashion as the sum of a
vector in S and a vector in T. In this case we say that the subspace T is a
supplement of the subspace S, and conversely. Given any subspace S of a
vector space V, there always exists a subspace Tsuch that V = S 7", but T
is not unique. Thus every subspace has a supplement, but unlike the situation
regarding complementation in set theory, supplements are not uniquely
determined.
Since the two definitions of direct sum are different, their properties
may also be expected to be slightly different. Note that the internal direct sum
is only defined for subspaces of a given vector space having zero intersection.
The external direct sum on the other hand is defined for an arbitrary pair of
vector spaces. Both and 4- satisfy commutative and associative laws, but
they are slightly different in detail. For example, although Sl S2 is equal
to S2 Si, the spaces V + V2 and V2 + V{ are in general not equal, but only
isomorphic. For vector spaces, a homomorphism is any linear mapping, while
an isomorphism is a one-to-one onto linear mapping. If Fis the internal direct
sum of two subspaces Sj and S2, then V is isomorphic to the external direct
sum of S1 and S2. Two vector spaces are isomorphic if and only if their
dimensions are equal. Therefore, the isomorphism between internal and
external direct sums implies that the dimension of the internal direct sum of
two subspaces equals the sum of their dimensions.
1.21 THE LATTICE OF IDEALS OF A LIE ALGEBRA
For the structure and classification of Lie algebras, the concepts of

subalgebra and ideal play the same fundamental roles that subgroups and
normal subgroups play in Lie group theory. If A and B are subspaces of a Lie
algebra, we denote by [A, B] the subspace spanned by all vectors [a, b], where
a e A and b e B. A subalgebra S of a Lie algebra L is a subspace which is
closed under the Lie multiplication, that is, which satisfies [S, S] c S. An
ideal I of a Lie algebra L is a subalgebra such that the Lie product of an
element of L with any element of / is in the subalgebra /, that is, for all x L
and all y e / , we have [x, y] /. Thus, an ideal / of a Lie algebra L may be
defined as a subspace which satisfies [L, /] <= /. As may be expected, sub-
algebras and ideals figure in the fundamental homomorphism theorems for
Lie algebras. In particular, the kernel of any Lie algebra homomorphism is an

ideal, while the image is a subalgebra.
The sum and intersection of ideals of a Lie algebra are again ideals, and
the ideals of a Lie algebra form a lattice under these two operations. In
addition, it follows from the Jacobi identity that the Lie product [7ls I2] of
two ideals is again an ideal. The situation regarding subalgebras is a little bit
different since the sum of two subalgebras need not be a subalgebra, although
the intersection of any set of subalgebras is still a subalgebra. Nevertheless,
the subalgebras of a Lie algebra still form a lattice. For both the lattice of
ideals and the lattice of subalgebras, the partial ordering is inclusion and the
greatest lower bound is intersection. In the lattice of ideals, the least upper
bound for a pair of ideals is the sum of those ideals. In the lattice of sub-
algebras, the least upper bound for two subalgebras is not their sum, but
rather the intersection of all subalgebras containing their union. Just as in the
case of finite groups, the structure of a Lie algebra can be studied by investi-
gating the properties of its lattice of subalgebras [22], [37], [101], [105], [141],
[142].
1.22 THE LEVI DECOMPOSITION OF A LIE ALGEBRA
The study of Lie algebras in general can be reduced to the study of two
special classes of Lie algebras: solvable Lie algebras and semisimple Lie
algebras. These types of Lie algebras may be defined in terms of ideals as
follows : With any Lie algebra L, there is associated a derived series of ideals
defined recursively by
and
for any positive integer k. It is readily proved by induction on k that L(k) is an

ideal of L and that the inclusion L (k+1) c L(k) holds. We say that L is a
solvable Lie algebra if the derived series of ideals
is eventually zero. Since a Lie algebra L is Abelian if and only if [L, L] = 0,

it follows that every Abelian Lie algebra is solvable. The simplest solvable
Lie algebra which is not Abelian is the two-dimensional affine Lie algebra.
If we choose a basis e^ , e2 for this Lie algebra such that
then L' consists of the multiples of e: , and L" = 0, proving that L is solvable.
In any Lie algebra, the solvable ideals form a sublattice of the lattice of
all its ideals, because the sum and intersection of solvable ideals are again
1.23 Semisimple Lie Algebras 45
solvable ideals. In particular, the sum of all the solvable ideals in a Lie
algebra is its unique maximal solvable ideal, called the radical of the Lie
algebra. We may define a semisimple Lie algebra as a Lie algebra which has
no Abelian ideals, other than 0. It is easy to see that a semisimple Lie algebra
cannot have any solvable ideals either, and hence a Lie algebra is semisimple
if and only if its radical is zero.
According to the Levi theorem [151], [245], any Lie algebra can be de-
composed as the direct sum, in the sense of vector spaces, of its radical R
and a semisimple subalgebra S:
That is, each element of the Lie algebra can be written uniquely as the sum
of an element in the radical and an element of this semisimple subalgebra.
Although the radical is uniquely determined, the Levi decomposition is not
unique because there may be several suitable semisimple subalgebras. In fact,
the Levi decomposition L = R S is unique only in the trivial case when S is
an ideal in L. Malcev, Goto and Harish-Chandra showed that if there are
two Levi decompositions, then there is an automorphism of the whole Lie
algebra which carries the one semisimple subalgebra onto the other [104],
[117], [160]. The problem of studying the structure of a Lie algebra is thus
almost reduced to studying semisimple and solvable Lie algebras.
The only question of structure remaining is to characterize the action of
the semisimple subalgebra S on the radical R. To investigate this point, we
need to use the methods of representation theory which we are about to
discuss. Anticipating a little, we can say that this question boils down to
knowing all the modules over a semisimple Lie algebra. Indeed, the condition
[S, R] c: R is equivalent to the statement that we may regard the radical R
as a module over the semisimple Lie algebra S.
1.23 SEMISIMPLE LIE ALGEBRAS
To study the structure of semisimple Lie algebras, one uses the fact that
they can be decomposed in terms of simple Lie algebras. A simple Lie algebra
is a non-Abelian Lie algebra which has no proper ideals at all. Every simple
Lie algebra is, as one would expect, semisimple, for the only nonzero ideal is
the whole algebra; if L' = L, this is not solvable, while if L' = 0, then L is
Abelian and consequently not simple. An example of a simple Lie algebra is
the real Lie algebra so(3, R), the familiar three-dimensional vector space R 3
equipped with the usual vector cross product. To see this, we must argue
that there are no proper ideals in the Lie algebra so(3, IR). Such an ideal
would be a proper subspace S of R3, a line or a plane passing through the
origin such that the vector cross product x x y of any vector x in S by a
vector y in R 3 must lie in S. Since the cross product of two vectors is
perpendicular to both and is nonzero unless the two vectors are collinear or
at least one is zero, no line or plane can have this property.
The fact that the rotation group S0(3, R) is simple (has no proper
normal subgroups) almost follows from the simplicity of its Lie algebra.
Since the closure of a normal subgroup is again normal and since the Lie
algebra of a closed normal subgroup of S0(3, R) is an ideal of its Lie algebra,
we only have to rule out discrete and dense normal subgroups. A normal
subgroup of S0(3, R) is dense if its closure is all of S0(3, R). If g is an element
of any normal subgroup S of S0(3, R), then all elements hgh~l which are
conjugate to g must also be in S. But this is equivalent to saying that every
rotation about any axis by an angle equal to that of the rotation g about its
axis must be in S. If g is not the identity, multiplying it by the inverse of a
rotation /ig/z"1 about an axis which has been shifted a small amount, we
obtain an element ghg~1h~l in S which is not the identity, but as close to it
as we wish. In this way one can further argue that any proper normal sub-
group contains a whole neighborhood of the identity element. Since any
neighborhood of the identity generates the whole rotation group, it follows
that SO(3, R) cannot have any proper normal subgroups at all.
Any semisimple Lie algebra S can be written as a direct sum of simple
ideals,
The elements of distinct S, must commute since a Lie product of elements

from distinct S, must be in their intersection which is zero. Hence, the structure
of a semisimple Lie algebra is completely determined by the structure of its
simple ideals. Combining these results with Levi's theorem allows us to
reduce any Lie algebra to simple and solvable Lie algebras. Thus any Lie
algebra L may be written as
where R is a solvable ideal and S t , , Sk are simple. However, it should be

noted that the S, are ideals of S, but not necessarily ideals of L.
The above theory can be best illustrated by an elementary example.
Perhaps the most familiar example of a Levi decomposition is that of the
Lie algebra of the group of all Euclidean motions. In this case the radical is
not only solvable, but even Abelian, and the semisimple subalgebra in the
Levi decomposition is simple. The radical consists of the subalgebra cor-
responding to translations or parallel displacements of R3. The simple
subalgebra is isomorphic to so(3, R) and corresponds geometrically to
rotations about some fixed point of R3. The nonuniqueness of the Levi
decomposition in this example comes from the arbitrariness of the choice of
the fixed point for these rotations.
1.24 The Baker-Campbell-Hausdorff Formula 47
1.24 THE BAKER-CAMPBELL-HAUSDORFF FORMULA
As an application of the Levi structure theorem, we shall discuss some

applications to the theory of differential equations. We have of course already
discussed the original work of Lie on differential equations. Recent work on
the application of Lie groups and Lie algebras to differential equations has
dealt with problems somewhat different from those studied by Sophus Lie.
One example is the study of equations of the type
where U and A are operators and U(0) = 1 is the identity. This equation
comes up in quantum mechanics as the Schrodinger equation for the
evolution operator U(t) of a system whose Hamiltonian operator is \(t),
except for a constant factor. Magnus gave a formal solution
for this problem, which converges in some interval about t = 0 if the operators
are finite matrices and A is a continuous function of t (see [159]).
Wichmann and Norman and Wei treated the case
where the constant operators \1 , , Xn span a finite-s

algebra [184], [185], [236], [246]. This situation arises in certain problems
of control theory [150]. Here the functions a,-(t), known as scalar controls, are
piecewise constant functions of time rather than continuous functions. In
control theory, the operators U(t) act on the states of the control system, and
describe how the states are transformed as a function of time. In this case,
the solution can be given locally in the form treated by Magnus as
Norman and Wei show that a solution can also be written as a product
These authors also studied the conditions under which the solutions converge
globally, that is, for all t.
In control theory, one frequently wants to know whether a system can be
adequately steered by a given set of controls [139]. In general, the states that
can be reached after a finite length of time from a given initial state under a
suitable class of controls form a certain submanifold of the state space. A sys-
tem is completely controllable if for any pair of states, there is a time interval
and a control on that interval such that the system trajectory, starting in the
one state, and steered by the control, ends at the other. The isotropy subgroup
of a given state consists of those control processes which, when applied to the
system initially in the given state, at the end return the system to that same
state. If a system is completely controllable, the state space itself can be
identified with the coset space of the group of all control processes modulo
the isotropy subgroup of any fixed state [40] . Kucera and Elliott have shown
that a system is completely controllable if it is controllable locally in a certain
sense [85], [86], [149], [150].
The point of departure for much of this work is a pair of formulas of
Baker and Hausdorff which give log(e* ey) and e~xyex in terms of com-
mutators of x and y. The first few terms of the expansions are
and
The omitted terms in both of these expansions involve successively higher

and higher order commutators of x and y. A computer program to develop
the coefficients of the Baker-Hausdorff formula has been written recently
[107] . The formula for log (e* ey) in general may fail to converge even in the
simplest case when x and y are finite matrices [16], [44], [118]. Magnus gives
a formula for l(t) by developing a continuous analogue of the Baker-
Hausdorff formula for log (ex ey), while Norman and Wei use the formula for
e~xy ex to obtain differential equations for the functions gfc). Conditions for
the solutions to be global can sometimes be stated in terms of the structure
of the Lie algebra generated by \(f).
Wichmann has utilized the Levi decomposition of finite-dimensional
Lie algebras to show that the problem of solving the differential equation
can actually be reduced to the case where the Lie algebra generated by \(t) is
simple. The Lie algebra generated by A(0 can be decomposed as the direct
sum of its radical and a semisimple subalgebra. Wichmann decomposes
A(t) as
where A 1 is in the radical and A2 is in the semisimple subalgebra. The original

problem may then be reduced to solving two simpler problems. The first of
these problems is that of solving
where A 2 (f) generates a semisimple Lie algebra. The second problem is that
of solving
also subject to the initial condition UÔ) = 1. If one can find a solution of
the first problem, then it is shown that the second problem can be reduced to
1.24 The Baker-Campbell-Hausdorff Formula 49
quadratures. The net effect is thus to reduce the case of a general Lie algebra
to the case of a semisimple Lie algebra. The semisimple subalgebra can be
further decomposed as the direct sum of its simple ideals. Thus we may
decompose A2 as
where the A 2j belong to those simple ideals, and hence mutually commute.
The first problem may then be shown to reduce to solving the problems
Chen has considered another problem,
which can be regarded as the problem of studying a flow with a variable

velocity field [52], [53], [122]. An infinitely differentiable vector field can be
regarded as an operator on the infinitely differentiable scalar functions by
identifying its direction with the directional derivative. Thus we identify the
vector field U with its dot product U V with the gradient operator. The set of
such operators forms a Lie algebra under commutation, and therefore by
this identification we can also regard the vector fields themselves as forming a
Lie algebra. That is, the Lie bracket of two vector fields is the vector field
We consider here V(t, x) not as a single vector field, but as a one-parameter

family of vector fields, indexed by the time t. This family of vector fields
generates a certain Lie algebra of vector fields under the Lie bracket com-
position. Chen shows that if the Lie algebra generated in this manner is
finite-dimensional, then the Levi structure theorem can again be used to
obtain an analogue of Wichmann's decomposition for the flow problem.
This depends basically on developing a more general transformation playing
the role of the similarity transformation appearing in Wichmann's differential
equation for A x .
CHAPTER 2
REPRESENTATION THEORY
2.1 LIE GROUP REPRESENTATIONS
One of the main tools in applications of the Lie theory is the concept
of a representation. A representation of a group is a homomorphism of the
group into a group of linear operators on a vector space. In many applica-
tions it is sufficient to treat representations in a fairly loose manner. For
example, one speaks of vectors, tensors, pseudoscalars, spinors and the like
as being various types of geometrical objects, such as directed arrows,
ellipsoids, and so on [39], [208]. Representations of the rotation group crop
up throughout mathematical physics in the form of spherical harmonics and
Legendre functions, multipole expansions and so forth. The actual representa-
tion concept is often held in the background. We must of course make this
intuitive conception more precise, and moreover we want to make our
discussion apply to any Lie group, not just to the rotation group. However,
we shall restrict our attention to analytic homomorphisms, and we assume
that all vector spaces have finite dimension.
We define a finite-dimensional representation of a Lie group to be an
analytic homomorphism of the Lie group into the general linear group of
invertible linear operators on a finite-dimensional vector space. By fixing
a basis in the vector space, we obtain a corresponding homomorphism onto
a matrix group. Since analytic homomorphisms of Lie groups induce
homomorphisms of the corresponding Lie algebras, every finite-dimensional
representation of a Lie group induces a representation of its Lie algebra.
A representation of a Lie algebra L consists of a vector space V and a
homomorphism /from L into the Lie algebra of all linear operators on V. The
term "representation" strictly refers to the pair (K/), but colloquially it is
often used to refer just to the homomorphism / alone. The requirement
that /be a homomorphism means that if x is in L, then/(x) is a linear operator
on K and/(x) depends on x linearly:
for all in L and all scalars c^ and c2 . Also we must have
51
52 Representation Theory 2.1
The correspondence between representations of Lie groups and those of

Lie algebras is not one-to-one, however, because a given representation of
the Lie algebra of a Lie group need not in general correspond to any rep-
resentation of the Lie group. Nevertheless, the major portion of work in-
volved in studying the group representations is frequently done when one
has found those of the Lie algebra.
A simple example of a Lie algebra representation which arises naturally
is the adjoint representation of any Lie algebra L. Each element x in the
Lie algebra L gives rise to a linear operator ad x acting in L, defined for
all y in L by
The operator ad x behaves just like a first order differential operator in that
it obeys the Leibniz rule for operating on a product,
Although this Leibniz rule is merely another way of writing the Jacobi
identity, this analogy with differential operators is frequently useful in Lie
algebraic computations. We note parenthetically that we may use the
operator ad x to rewrite one of the symbolic Baker-Hausdorff formulas as
The other Baker-Hausdorff formula may likewise be written as
where
The mapping ad which takes x into ad x is a linear mapping from the

Lie algebra L into the space of linear operators on L. To show that it is a
representation we need only verify that it preserves Lie products,
But for any element z in the Lie algebra,
Using the Jacobi identity and antisymmetry, we have
so that ad is indeed a Lie algebra homomorphism. The pair (L, ad) is then a
representation of the Lie algebra L, called the adjoint representation of the
Lie algebra L.
2.2 Modules over Lie Algebras 53
The adjoint representation can be used to define a certain bilinear form,

due to Killing, which is of great importance for the structure theory of
Lie algebras. The Killing form is a symmetric bilinear form defined for any
finite-dimensional Lie algebra L by
The Killing form permits geometrical ideas such as orthogonality to be used

in the study of a Lie algebra. The Killing form is invariant under every
automorphism a of the Lie algebra :
The importance of the Killing form derives in part from the fact that a
Lie algebra is semisimple if and only if its Killing form is nonsingular.
Solvability can also be characterized in a simple way in terms of the Killing
form. A Lie algebra L is solvable if and only if the Killing form is zero on the
derived algebra L' = [L, L] , that is, if (x, y) = 0 for all x and y in L'.
A linear Lie algebra is a Lie algebra whose elements are linear operators
in some vector space, and whose Lie product is the commutator. A representa-
tion is faithful if it is a one-to-one mapping. If a Lie algebra has a faithful
representation, then the Lie algebra is isomorphic to the image of the
representation, which is a linear Lie algebra. For a semisimple Lie algebra,
the adjoint representation is faithful. For if x belongs to the kernel of ad, then
for any y in L we have
and hence this kernel is an Abelian ideal of L. Consequently if L is semisimple,

then this Abelian ideal in L must be zero, and hence, the adjoint representa-
tion is faithful. But the adjoint representation need not be faithful for Lie
algebras in general. Ado proved that every finite-dimensional Lie algebra
has a faithful finite-dimensional representation [2] . Therefore, every finite-
dimensional Lie algebra is isomorphic to a Lie algebra of linear operators,
and hence, choosing a basis, to a Lie algebra of matrices. One of the ideas
involved in the proof of the Ado theorem is the use of the Levi decomposition
to reduce the theorem to the case of a solvable Lie algebra. For a solvable Lie
algebra one can prove the Ado theorem by induction on the dimension.
2.2 MODULES OVER LIE ALGEBRAS
In some respects the concept of a representation is unnecessarily com-

plicated. The object being represented, Lie group or Lie algebra, is first
mapped by a homomorphism into a similar object composed of linear
operators which in turn act on a vector space. There is no loss of generality,
of course, if we regard the elements of the original object as affecting the
linear transformations on the vector space directly, disregarding the
intermediate homomorphism. When we take this viewpoint and the object

being represented is a group, we speak of the vector space as a module
over the group. When we are dealing with an algebra, we refer to the vector
space as a module over the algebra.
A module over a Lie algebra L consists of a vector space M and a bilinear
mapping L x M - M, called the module product rule, such that if the element
in M assigned to x in L and v in M is denoted by xv, then
for all Xj and x2 in L and all v in M. As usual, we can feel free to replace the
bilinear mapping L x M -* M by an equivalent linear mapping L M - M,
if we prefer to do so.
The study of modules over an algebra is really the same as the study of
the representations, the only difference being a point of emphasis. If M is a
module over a Lie algebra L, then for any x in L, let/(x) be the linear operator
on M which maps v in M into the vector xv in M. Then (M,/) is a representa-
tion of the Lie algebra L. Conversely, given a representation (M, /) we can
make the vector space M into a module by defining the product xv of x in L
and v in M to be/(x)y. Thus the concepts of module and representation are
equivalent. In talking about modules we emphasize the vector spaces in-
volved, whereas in talking about representations, we emphasize the homo-
morphisms. Generally speaking, the module formulation is the easier to work
with.
Representations and modules can be defined for Lie groups as well as
for Lie algebras. An analytic representation of a Lie group G is a pair con-
sisting of a vector space M and an analytic homomorphism/of G into the Lie
group GL(M) of invertible linear operators on M. If n is the dimension of M,
then GL(M) is isomorphic to the Lie group GL(n, C) of nonsingular n x. n
matrices over C. Hence, a representation of a Lie group corresponds to a
matrix representation, that is, a homomorphism of the Lie group into the
matrix group GL(n, C).
We may also define a module M over an associative algebra A in much
the same way that we defined a module over a Lie algebra. The main dif-
ference is that the requirement
for Lie modules is replaced by the requirement
for all x and y in A and for all v in M. If the associative algebra A has a unity
element e, then we may also want to demand that
hold for all t; in the module M.

2.2 Modules over Lie Algebras 55
Lie modules have many applications within Lie algebra theory. For
example, the adjoint representation is equivalent to regarding any Lie
algebra as a module over itself, the module product being taken to be the
Lie product itself. By a slight generalization of this idea, we may also regard
any ideal of a Lie algebra L as a module over L. Again, any extension of a Lie
algebra L, that is, any Lie algebra containing L as a subalgebra, may be
regarded as a module over L. We have already seen that modules enter into
the structure of Lie algebras via the Levi decomposition. In this case the
radical is a module over the semisimple subalgebra, and the module product
is again the Lie product itself. Lie modules also are useful in the further study
of the structure of semisimple Lie algebras via the theory of roots. In this
case the whole Lie algebra is regarded as a module over a certain Abelian
subalgebra, the Cartan subalgebra. We shall come back to this in 2.7.
Any representation (M,/) of a Lie algebra L can be replaced by a
matrix representation. If a basis elt , enis selected in the vector space M,
then the linear operators /(x) correspond to matrices [/'/x)] defined by
The mapping which assigns the matrix [/'/*)] to x is called a matrix rep-
resentation of the Lie algebra L. If a different basis is selected for M, of
course, the matrices of the matrix representation will all be subjected to a
common similarity transformation. Matrix representations related by a
common similarity transformation are considered to be equivalent or
isomorphic. Equivalence of Lie group representations is defined just as for
Lie algebras. Two matrix representations of a Lie group are equivalent or
isomorphic if they are related by a common similarity transformation.
A homomorphism h of one module M over a Lie algebra L into another
module N over L is a linear transformation h : M -> N which preserves the
multiplication by elements of L, that is,
for x in L and v in M. If the homomorphism h is one-to-one and onto, and

hence invertible,then we call it an isomorphism, and we say that the modules M
and N are isomorphic to each other. If /is the representation of L associated
with the module M and if /' is the representation of L associated with the
isomorphic module N, then we have hf(x) = f'(x)h, and if h is invertible,
we may write this as
Such representations, corresponding to isomorphic modules, are called

equivalent representations. If M and N are isomorphic modules, then there
exists a choice of bases for both modules such that the representation matrices
thus obtained will be the same for the two modules.
If a module M over a Lie algebra L has a subspace S which remains

invariant under the action of L, symbolically LS c S, then S itself can be
regarded as a module over L, and we call it a submodule of the module M.
That is, a subspace S of a module M is a submodule if xv is in S for all x
in L and all v in S. In the physics literature modules are also called rep-
resentation spaces, while submodules are called invariant subspaces. The
submodules of a Lie module form a lattice under intersection and sum. A
module which has no submodules other than zero and itself is called an
irreducible module, or a simple module, and the corresponding representation
is then also called irreducible. In other words, a representation is irreducible
when the representation space has no nontrivial invariant subspace.
Since a module M is a fortiori an Abelian group, and a submodule N
is a normal subgroup, we can form the quotient group M/N whose elements
are the cosets v + N, with v in M. Since N is a vector space, multiplication by
complex numbers is well-defined on M/N, and hence M/N is a vector space.
Similarly, since N is a module over L, multiplication by elements of L is
well-defined on M/N, and hence M/N may be considered as a module over
the Lie algebra L, called the quotient module of M with respect to N. It is
easily shown that if h is a homomorphism of a module M into a module N,
then the kernel K of h is a submodule of M, and the quotient module M/K
is isomorphic with the image of the homomorphism, which is a submodule
of N.
2.3 DIRECT SUM DECOMPOSITIONS OF LIE MODULES
There are several useful ways of combining modules to obtain new ones,
the most immediate of these being the direct sum. Since modules over Lie
algebras are vector spaces, it is meaningful to talk about their internal and
external direct sums as vector spaces. If M : and M2 are modules over a
Lie algebra L, their vector space external direct sum MJ + M2 can also be
regarded as a module over L by defining the action of L componentwise.
Explicitly, the action of any element x of the Lie algebra L on any vector
(v1, v2) in the external direct sum module Ml 4- M2 is given by
If a module M = M1 M2 is the internal direct sum as a vector space of

these submodules M\ and M 2 , then M is isomorphic as a module to the
external direct sum module Ml + M2.
If a module M is the direct sum of two nonzero submodules, we say that
M is decomposable, and the corresponding representation is also said to be
decomposable. Otherwise, M is called indecomposable. Every finite-dimen-
sional module which is decomposable is a direct sum of indecomposable
modules. For the stu'l^ of finite-dimensional modules over Lie algebras, it
2.4 Lie Module Tensor Product 57
suffices to consider the indecomposable modules. Obviously an irreducible

module must be indecomposable, but in general an indecomposable module
need not be irreducible.
A module which is either irreducible or is the direct sum of irreducible
submodules is called a completely reducible module or a semisimple module.
A necessary and sufficient condition for a finite-dimensional module to be
completely reducible is that every submodule be a direct summand, that is,
for every submodule N there is another submodule AT such that
Every finite-dimensional module over a semisimple Lie algebra is completely

reducible. Our attention can thus be directed to a study of the irreducible
modules of a semisimple Lie algebra since these form the building blocks
from which any other module can be constructed. However, this statement
does not hold for Lie algebras in general.
Complete reducibility also holds for modules over compact Lie groups,
which simplifies their representation theory substantially [124], [192]. For
any representation of a compact group by linear operators on a Hilbert
space, the inner product on the Hilbert space can be replaced by another
inner product with respect to which the operators are unitary [80], [181].
Thus, only unitary representations, i.e., representations by unitary linear
transformations, need be considered when one is dealing with compact
groups. We recall that unitary representations are also of interest in quantum
mechanics in view of Wigner's unitary-antiunitary theorem. For this
application we must of course also treat ray representations, but these can
be replaced by ordinary unitary representations of the universal covering
group [18]. Every unitary representation of a compact group is a direct sum
of irreducible unitary representations acting on mutually orthogonal sub-
spaces. Moreover, the irreducible unitary representations of a compact group
are all finite-dimensional [144], [180]. Thus the representation theory of a
compact group on a Hilbert space reduces to the study of irreducible unitary
representations on finite-dimensional vector spaces. In particular, the rota-
tion group SO(3, R) and its universal covering group SU(2), being compact,
can only have finite-dimensional irreducible representations. For a non-
compact group there may not exist any nontrivial finite-dimensional unitary
representations, so we must also study the infinite-dimensional ones [156].
2.4 LIE MODULE TENSOR PRODUCT
A second useful way of combining Lie modules which arises frequently

in applications is by means of the tensor product. Since Lie modules are
vector spaces, their tensor products are again vector spaces. The tensor
product vector space M x M2 of modules Ml and M2 over a Lie group G

can be considered as a module over G by defining
We can pass from modules over Lie groups to modules over Lie algebras by
differentiation, or vice versa by exponentiation. To motivate this discussion
we shall restrict our attention to linear Lie groups and algebras, that is,
groups and algebras whose elements are linear operators in a finite-dimen-
sional vector space. If x is an element of a linear Lie algebra L, then g = etx
= 1 4- tx + is an element of a linear Lie group. For sufficiently small real
numbers t, we could ignore higher order terms. Then, inserting this expan-
sion for g in the above definition of its action on the tensor product space,
and comparing powers of f, we arrive at the formula
We may now drop the restriction to linear Lie algebras by using this
formula to define tensor products of modules over abstract Lie algebras. If
M j and M 2 are modules over a Lie algebra L, this formula defines an action
of L on the tensor product vector space Ml M2 making it into a module
over L. It can be verified directly that the axioms for a module over a Lie
algebra are satisfied without reference to Lie groups. As a mnemonic aid
we may further note that the rule
may be viewed as an abstract generalization of the Leibniz rule for differ-

entiating a product.
Consider the tensor product of two modules over a Lie algebra. If the
Lie algebra is semisimple, then the original two modules are completely
reducible, and so can be written as a direct sum of irreducible submodules.
The tensor product is then just the direct sum of all the tensor products of
these irreducible submodules. To reduce the tensor product of arbitrary
modules over a semisimple Lie algebra, therefore, it is sufficient to study the
reductions of tensor products of irreducible modules as direct sums of irre-
ducible modules. In the applied literature, these expansions of tensor prod-
ucts of irreducible submodules as direct sums of irreducible submodules are
called Clebsch-Gordan series. In particular, the Clebsch-Gordan series for
the rotation group play a fundamental role in the quantum mechanical
description of atomic and nuclear spectroscopy.
Another module operation closely related to the tensor product is the
dual operation. The dual vector space M* consisting of all scalar-valued
2.5 Tensor and Exterior Algebras 59
linear functions on a module M over a Lie algebra L may be considered as

a module over L. If the action of L on M* is defined by
for all 4> in M* and all m in M, then straightforward computation verifies

that M* satisfies the module axioms. This formula may be understood in
terms of the natural pairing of the space M * with M, that is, the linear
mapping of M* M into the base field which takes 4> m to (j>(m) for all
(f) in M* and all m in M. The definition of the dual module M* just amounts
to the requirement that this natural pairing, or contraction, be a Lie module
homomorphism, if we take the base field as a trivial module over L. If a basis
{et} is chosen in M, and the corresponding dual basis {e1} is chosen in M *,
then the representation matrices in the dual module are the negative trans-
poses of the representation matrices in the original module. The action of
x on the basis elements e can be written as
Then, using a simple computation shows that
Thus, here DJ,-(x) is the matrix representative associated with the module M,
and the negative transpose D'/x) gives the corresponding matrix repre-
sentation associated with the dual module M*.
2.5 TENSOR AND EXTERIOR ALGEBRAS
Another concept from vector space theory having important applica-

tions in the theory of Lie algebras is the tensor algebra of a vector space.
Although this algebra is infinite-dimensional, it has extremely simple prop-
erties. The tensor algebra is an associative algebra which behaves in many
respects like a polynomial algebra except that the tensor algebra is not
commutative. For example, each element can be assigned a degree, having
the expected properties, and one says that the tensor algebra is a graded
algebra [56].
We define the tensor algebra by combining the ideas of tensor products
and external direct sums. The construction given for the tensor product of
two vector spaces can obviously be extended to define a tensor product of
any finite number of vector spaces. In particular, the tensor product of V
with itself r times, called the r-th tensor power,
is the space of tensors of rank r. Strictly speaking, the tensor products

r + s V and ((X)r K) ((x)s V) represent different vector spaces [157], but
we shall identify them when we construct the tensor algebra on K The

tensor product of a tensor in (x)r V and a tensor in (X)s V may be regarded
as a tensor in (X) r+s V by making the natural identification, for instance, of
with
This suggests the idea of forming the contravariant tensor algebra T(V)
on a vector space Kover a field F. The tensor algebra is defined as the infinite
weak external direct sum
of all the tensor powers of the vector space. The infinite sum here is weak
in the sense that we consider only finite linear combinations of the elements
in the spaces F, V, V V, , so that no actual infinite summations ever
occur, and no questions of convergence can arise. Being a direct sum of
vector spaces, the tensor algebra is clearly a vector space, and with the
tensoring together operation as multiplication, T(V) is an associative
algebra with unity. If V is an n-dimensional vector space, then each tensor
power (X)r Fhas dimension nr, and hence the tensor algebra T(V) is infinite-
dimensional.
Another algebra, closely related to the tensor algebra, is obtained by
antisymmetrizing the tensor product. The resulting theory of the exterior
algebra provides an elegant way to treat determinants, and has important
applications to the theory of Lie groups and Lie algebras. The antisym-
metrization process can be effected most quickly by means of a quotient
construction as follows.
A two-sided ideal I in an associative algebra A is a subspace which
satisfies IA e / and AI c /. The quotient algebra of an associative algebra A
with respect to a two-sided ideal / is the quotient vector space under the
multiplication
Starting with the tensor algebra T(V) of a vector space V, which is an

associative algebra under the (x) operation, consider the two-sided ideal A
generated by all elements of the form x x, where x e V. The quotient
algebra
is an associative algebra with unity called the exterior algebra or the Grass-
mann algebra of the vector space V. It is customary to use the symbol A
to denote the multiplication operation induced by (g) in the exterior algebra
E(V), so that we have
2.5 Tensor and Exterior Algebras 61
for the cosets of any tensors tl,t2e T(V). If S t and 52 are any two sub-
spaces of the exterior algebra E(V), then S1 A S2 denotes the subspace of
all linear combinations of elements s1 A s 2 , where slSl and s 2 eS 2 .
Since E(V) is an associative algebra, we have
for any three subspaces of the exterior algebra. The vector space Kis naturally
embedded in the exterior algebra E(V), just as it is in the tensor algebra, and
we may identify the vector x e V with the coset x + A e E(V}.
The exterior power
(r copies)
is a subspace of the exterior algebra. Elements of the exterior power /\rV
are sometimes called r-vectors. The whole exterior algebra E(V) is the direct
sum of the exterior powers AT Thus the exterior algebra can be broken up
into pieces each with its own degree, and it is also a graded algebra. The
grading of the exterior algebra is directly related to the corresponding
grading of the tensor algebra. The ideal A is the direct sum of the individual
pieces Ar A fl ((X)T) lying in each tensor power, and the quotient space
((X)T)//T is naturally isomorphic to the exterior power AT. The exterior
square V A V of a vector space V is a vector space generated by exterior
products x A y of vectors x, y in V. If x V, then
since x x belongs to the ideal A by definition. If x and y are in V, then
and writing this out and using
we find that
Thus, the exterior product x A y of two vectors x and y in V is bilinear in

x and y and is antisymmetric under their interchange.
If V is an n-dimensional vector space with basis elt , en, then the
exterior square V A V is a vector space of dimension
with basis ei A ejy where 1 ^ i < j ^ n. More generally, the products
eh A A ejr with strictly increasing indices j^ < < jr form a basis for
the exterior power space f\rV since any change in the order of factors in a
product at most changes its sign. If V is an n-dimensional vector space, the
exterior power /\rV has dimension n and
zero for r > n, and hence the exterior algebra E(V) has the finite dimension 2".
We now indicate some of the many interesting applications of exterior

algebras within vector space theory itself. The vectors xl , , xr are linearly
independent if and only if their exterior product
is not zero. Any r-vector of the form
where x t , , xr e K is called a simple r-vector. If S is an r-dimensional sub-

space of a vector space K, then there exists a simple r-vector se AT such
that the condition for a vector x to belong to S is s A x = 0. If ev , , er
is a basis for the subspace S, then we can take s = el A A er.
The exterior square V A V may be identified with the subspace of
V (x) V consisting of all linear combinations of elements of the form
where x and y are in V. More generally, the exterior power f\kV may be
identified with the subspace of (x)fcK consisting of all linear combinations
of elements of the form
where the x,- are vectors in V, and the sum goes over all permutations n of
the indices 1, , k. The sign factor <5n here depends on the parity of the
permutation. It is +1 for even permutations and 1 for odd permutations.
If a is a linear mapping from a vector space Vl into a vector space V2 , then
is a linear mapping from (^)rV1 to 'V2. Since antisymmetric tensors

remain antisymmetric under (X)ra, we obtain an induced linear mapping
Ara from f\rV^ to /\rV2. If V is n-dimensional, then /\"V is one-dimensional.
The operator A"a induced by any linear operator a on V may be identified
with the determinant of a. If el , - , en is a basis for V, then we have,
explicitly,
The exterior powers of a module are essential for the construction of

the irreducible module's of certain Lie algebras. It is clear that the tensor
power (x)rM of a Lie module M is again a module over the Lie algebra L.
The subspace Ar, spanned by all elements of the form
where m is in M and where ae(x) k M and be(x) r ~*~ 2 M are tensors of

lower rank (0 ^ k ^ r 2), is a submodule of the tensor power (x)rM. To
2.6 Universal Enveloping Algebra of a Lie Algebra 63
see this, we consider the action of an element x of the Lie algebra on

a m (x) m (g) b, and use
The exterior power, being the quotient space
is, therefore, also a module over the Lie algebra L. The action of an element
x of L on the elements of ArM is induced by the action of x in (X)rM, namely,
2.6 THE UNIVERSAL ENVELOPING ALGEBRA OF A LIE

ALGEBRA
Of central importance in Lie algebra theory is another algebra obtained

from a tensor algebra by means of a quotient construction. This associative
algebra, called the universal enveloping algebra of a Lie algebra, is important
because it allows us to translate questions about Lie algebras into corre-
sponding questions about associative algebras. We have seen that Lie group
theory reduces locally to the theory of Lie algebras. By means of the envel-
oping algebra, the theory of Lie algebras is reduced to the even better
understood theory of associative algebras. Homomorphisms of one Lie
algebra into another induce corresponding homomorphisms of their uni-
versal enveloping algebras. Moreover, the ideal generated in the universal
enveloping algebra by the kernel of a Lie algebra homomorphism is the
kernel of the induced homomorphism on the corresponding universal
enveloping algebras. More generally, Lie subalgebras and Lie algebra ideals
yield corresponding associative subalgebras and ideals in the universal
enveloping algebra.
Intuitively, the universal enveloping associative algebra U(L) consists
of all polynomials in elements of the Lie algebra L and a unity element,
with the Lie product [x, y] of elements x and y in L being identified with the
commutator xy yx. More formally, we can construct the universal
enveloping algebra as follows. Since the Lie algebra L is a vector space,
it is possible to construct its contravariant tensor algebra T(L) = F + L
+ (L (x) L) + , as before. In this associative algebra T(L\ we consider
the two-sided ideal K generated by the set of all elements of the form
where x and y are elements of L. The ideal K thus contains the differences
between Lie algebra products and the corresponding commutators in the
associative tensor algebra. If we consider the associative quotient algebra
U(L) = T(L)/K, then Lie products will not be distinguished from, com-
mutators since they belong to the same coset. The associative algebra
is called the universal enveloping algebra of the Lie algebra L. As with any
associative algebra, we can also make U(L) a Lie algebra using the com-
mutator operation as the Lie product. If we do this, we can consider L to be
injected homomorphically into U(L), considered as a Lie algebra.
The associative algebra U(L) is useful because of the following property.
Suppose that A is an arbitrary associative algebra, and that A is also given
the commutator Lie algebra structure. Any homomorphism of L into A,
considered as a Lie algebra, has a unique extension to an associative algebra
homomorphism of U(L) into A. Now a representation of a Lie algebra L is
a Lie homomorphism of L into the associative algebra of all linear operators
on the module, with the Lie product of two linear operators being their
commutator. Thus every representation of a Lie algebra L can be extended
to a representation of its universal enveloping associative algebra U(L), and
we see that every module over L can also be regarded as a module over its
enveloping algebra U(L). This idea is central to certain proofs of complete
reducibility for modules over semisimple Lie algebras which are based on
the universal enveloping algebra.
The associative algebra U(L) acts on the module M by letting
for all n, for x 1? , xn in L and m in M, and extending the definition to

all of U(L). It is easily checked that this definition is independent of the
choice of coset representatives and so actually defines an action of
U(L) = T(L)/K on M.
The enveloping algebra U(L) has a great deal of structure which we
may describe briefly here. If (xj is a basis for L, then the monomials of the
form
where we take the trivial monomial 1 in case n = 0, span T(L), and hence
their cosets span U(L). A result of Poincare, Birkhoff and Witt is that if we
only take monomials having their indices ij in ascending order, allowing
repetition, then the cosets of these monomials, again including 1, form a
basis for U(L).
It can be shown that U(L) has no nonzero zero divisors and that, for
finite-dimensional L, there cannot be an infinite sequence of ideals or one-
sided ideals in U(L) each properly contained in the next. If L is finite-
dimensional, then every ideal of L7(L) is finitely generated. From this, one
can conclude by the Goldie-Ore theorem that U(L) has a (two-sided) divi-
sion ring of fractions [61], [98], [99], [135]. A division ring D is said to be a
division ring of fractions for U(L) if U(L) is a subring of D and every element
of D can be written as a~ 1b and cd~ l, where a, b, c and d are elements of U(L).
2.7 Nilpotent and Cart an Subalgebras 65
The universal enveloping algebra has still another structure of con-

siderable importance [224]. The tensor product of two associative algebras
is an associative algebra under the product
In particular, then U(L) U(L) is an associative algebra and the mapping

L -* U(L) (g) U(L) which takes any element / L into the element 1 (g) /
+ / (x) 1 e U(L) (g) U(L) is a Lie algebra homomorphism. This Lie homo-
morphism can then be extended to an associative algebra homomorphism
which we call the diagonal mapping. The multiplication map A x A -> A

for any associative algebra A is a bilinear mapping which can of course be
replaced by a linear mapping A A - A. A linear mapping which goes in
the opposite direction, A -> A (g) A, is sometimes called a comultiplication,
and >1 may be called in this case a coalgebra. Thus we may regard the uni-
versal enveloping associative algebra U(L) to be both an algebra and a
coalgebra.
2.7 NILPOTENT AND CARTAN SUBALGEBRAS
We now discuss some ideas, many dating back to Cartan, needed to

analyze the structure of Lie algebras in more detail [45], [46]. Although
these results are mostly important for the structure of simple Lie algebras,
it is possible to apply some of these ideas to Lie algebras in general. A Lie
algebra L is said to be nilpotent if the lower central series of ideals
is eventually zero. We must distinguish this sequence of ideals from the

derived series of ideals L(k) used in defining solvable Lie algebras. Since it is
easy to show that L(k) <=. L2k, it follows that every nilpotent Lie algebra is
solvable. On the other hand, the two-dimensional non-Abelian Lie algebra
defined by [e^, e2] = el is solvable, but not nilpotent since Lk is spanned
by e1 for all k ^ 1. Nevertheless, solvable Lie algebras are only a little bit
more general than nilpotent Lie algebras. A finite-dimensional Lie algebra
L is solvable if and only if its derived algebra [L, L] is nilpotent. Engel showed
that a Lie algebra is nilpotent if and only if ad x is a nilpotent linear operator
(that is, some power of ad x vanishes) for every element x in the Lie algebra
[133]. More recently, Barnes showed that a finite-dimensional Lie algebra
is nilpotent if and only if all its maximal subalgebras are ideals [21].
We now study a certain special type of nilpotent subalgebra called a
Cartan subalgebra. The Fitting null component LQH of a Lie algebra L with
respect to a nilpotent subalgebra H is the subspace of L on which ad h is
nilpotent for all h E H. That is, an element y in L belongs to LH if for every

h H there is an integer n > 0 such that
Since H is a nilpotent subalgebra of L, Engel's theorem implies that

H c L# . A Cartan subalgebra H of a Lie algebra L is a nilpotent subalgebra
for which H = Lj. In other words, a Cartan subalgebra H of a Lie algebra L
is a nilpotent subalgebra such that if x e L satisfies
for all h in H, then x is also in H. There is an alternative characterization

of a Cartan subalgebra which is occasionally useful. A nilpotent subalgebra
H of a Lie algebra L is a Cartan subalgebra if and only iff/ is not contained
as an ideal in any subalgebra of L larger than H itself. Every finite-dimen-
sional Lie algebra contains at least one Cartan subalgebra [23] . In general,
a Lie algebra L may have more than one Cartan subalgebra, but they all
have the same dimension /, called the rank of the Lie algebra L. The Cartan
subalgebras of a semisimple Lie algebra are maximal Abelian subalgebras.
However, the reader should be warned that the converse is not true. A
maximal Abelian subalgebra of a semisimple Lie algebra need not be a
Cartan subalgebra.
2.8 WEIGHT SUBMODULES
A useful tool for the classification of modules over semisimple Lie

algebras is provided by the concept of a weight. The study of weights also
has a practical value for exhibiting the internal structure of a Lie module,
enabling one to see what that module contains.
Although the concept of weight is most useful in connection with semi-
simple complex Lie algebras, we can define weights for a module M over
any Lie algebra L. A linear form /i e L* is called a weight of M if there exists
a nonzero vector v in M such that
The vector v is thus an eigenvector for all elements of L simultaneously,

with n(x) being the eigenvalue corresponding to the operator x acting in M.
The weight n may thus be considered as this collection of eigenvalues. Lie
proved that if L is a solvable Lie algebra over the complex numbers, then
any nonzero module over L has at least one weight. The set of all weights
of a module is called its weight system. In the case of a single linear operator
acting on a finite-dimensional vector space over the complex number field,
there is a very detailed structure theory culminating in the Jordan canonical
form. In that theory it is necessary to generalize slightly the concept of
eigenvectors in order to obtain direct sum decompositions. For nilpotent
2.9 Roots of Semisimple Lie Algebras 67
Lie algebras over the complex number field we can develop a rather similar
albeit less detailed theory. To obtain direct sum decompositions of modules
over nilpotent Lie algebras, it is similarly necessary to generalize somewhat
the concept of a simultaneous eigenvector.
A nonzero vector v in M is called a weight vector, or generalized simulta-
neous eigenvector, if there exists an integer p such that
for all x in L. The set of all weight vectors corresponding to a given weight u,
together with the zero vector, is a submodule M of the module M, called
the weight submodule for the weight p. In the weight submodule M each
element x in L is represented by an operator which differs from a multiple
of the unit operator by a nilpotent operator. For a module over a nilpotent
Lie algebra we can find a basis consisting entirely of weight vectors. In other
words, a module over a nilpotent Lie algebra is the direct sum of its weight
submodules.
To apply these ideas to a semisimple complex Lie algebra L, we make
use of one of its Cartan subalgebras H. Any module M over the semisimple
Lie algebra L is also a module over the Abelian subalgebra H. Any module
over a semisimple Lie algebra L can then be written as the direct sum of its
weight submodules with respect to H :
In this particular application, the weight submodules actually contain only

ordinary eigenvectors.
2.9 ROOTS OF SEMISIMPLE LIE ALGEBRAS
Let H be a Cartan subalgebra of any Lie algebra L over the complex

numbers, and let H* denote its dual space, that is, the set of all complex-
valued linear forms on H. A linear form a e H * is called a root if there exists
a nonzero element x e L such that
for all h e H. Roots play the same role for the structure of Lie algebras that
weights play for Lie modules. The concept of root is in fact just a special
case of the concept of weight which applies to the adjoint representation of
the Lie algebra. Thus, regarding L itself as a module over H via the adjoint
representation, we obtain a decomposition of L as a direct sum of its root
spaces LaH. The Cartan subalgebra H itself is just the root space for a = 0
since H = L?H.
For semisimple Lie algebras over the complex numbers, the above
results may be further sharpened. The restriction of the bilinear Killing form
on L to the Cartan subalgebra H is in this case nonsingular. This means that

for every linear form a e H* there is a unique vector ha E H such that
for all h e H, and thus we can identify the dual space H* with H itself. The
root system has a number of special properties for semisimple Lie algebras.
For example, if a is a nonzero root, so is a, but there are no roots which
are multiples of a other than a, 0 and - a. Moreover, for semisimple complex
Lie algebras, the root spaces LaH corresponding to nonzero roots a are all
one-dimensional [81]. Thus, for any nonzero root a, there is, up to a factor,
a unique vector ea e L, called a root vector, such that
for all h e H. In other words, for all h in H, this root vector ea is a simultaneous
eigenvector of the linear operators ad h acting on L.
If ea and ef are root vectors, then by the Jacobi identity,
for all heH. If a + /? is a nonzero root, then it follows that [e a ,fy] is a

multiple of e a +0, while if a + ft is not a root, then [ea, e^\ = 0. Finally, in
the case a + /? = 0, we have the formula
A semisimple complex Lie algebra L is the direct sum of the /-dimensional

Cartan subalgebra H and all the one-dimensional root spaces LaH corre-
sponding to nonzero roots,
Thus, a basis for H together with the ea determines a basis for L. It follows,
therefore, that the total number of nonzero roots of a semisimple Lie algebra
is equal to its dimension minus its rank. If a and /? are roots such that
a + ft 96 0, then the corresponding root vectors ea and e^ are orthogonal.
Hence ea and e_ a are both null vectors, but they are never orthogonal to
each other, and hence the bilinear Killing form restricted to the plane
spanned by them is nonsingular.
To further study the root system of a complex semisimple Lie algebra,
we introduce a certain Euclidean space and an ordering of the roots. Let
H% denote the real /-dimensional subspace of H* whose elements are all
real linear combinations of roots. Since H* can be identified with H, we
may define a metric on the dual space H* by
2.9 Roots of Semisimple Lie Algebras 69
for all a and ft in H *. If a and ft are nonzero roots, then (a, ft) is real and (a, a)
is positive nonzero. The Killing form is thus positive definite on the real
space H|, making this a Euclidean space with the Killing form as inner
product. We obtain a total ordering of the roots by choosing an arbitrary
ordered basis y l 9 , yl for the real Euclidean space H.
where the components rv , , rl of the vector , are real numbers. We define

a lexicographic ordering in H% by saying that a vector is higher than a
vector ri, written > rj, if the first nonzero component of their difference
Y\ is positive. A root a is called a positive root if a > 0, a definition which
depends, of course, on which choice of lexicographic ordering is made. A
simple root is a positive root which is not the sum of two positive roots.
There are exactly / simple roots, a l 5 , a,, which form a basis for H*.
Each positive root is a linear combination with nonnegative integer coeffi-
cients of the simple roots, and each negative root is such a combination
with nonpositive integer coefficients. Thus, for any root a, we may write
where the integers k{ , , kt have the same sign ; if a > 0, then k1 ^ 0, ,

fc, ^ 0, while if a < 0, then /Cj ^ 0, , kt ^ 0. The sum fcj + + fc,,
called the /eue/ of the root a, is positive for a > 0, zero for a = 0, and negative
for a < 0.
A simple and useful property of weights and roots is their additivity
with respect to taking tensor products. If M^ and NVH are weight submodules
of the modules M and N over a nilpotent Lie algebra H, then the tensor
product Mfr NVH is contained in the weight submodule (M N)ft +v of
the module M N. There are vectors v and w in M and NVH, respectively,
such that
for all h H, and hence,
Thus, /* + v is one of the weights in the tensor product of these weight sub-
modules, and in fact it is the only weight in this tensor product.
In general, the weight system of the tensor product M N of any two
modules M and N is the collection of sums of a weight of M and a weight
of N. That is, the weight system of a tensor product of two modules can be
obtained from the weight systems of the two individual modules by a process
of vector addition. These results can be used as follows to clarify the action
of a Lie algebra L on one of its modules M. This action L x M -> M can
70 Representation Theory 2.1 0
be formally identified with a module homomorphism L M - M by using

the tensor product lifting property. It follows that the action of a root
subspace L*H on a weight submodule M is given by the shifting rule
LM c M+a. This shifting rule says that if a root vector ea acts on a
weight vector v with weight //, then eat> is either zero, or else a weight vector
corresponding to the shifted weight \i + a. We have in fact already met one
special case of this shifting rule before, namely,
which we proved directly, using the Jacobi identity.

Suppose now that M is a module over a semisimple Lie algebra L.
Acting on a weight submodule M repeatedly with ea and e_ a , we get weight
submodules corresponding to a whole ladder of weights (i + za, where
z = 0, 1, 2, . Since we are dealing with finite-dimensional modules,
all but finitely many of the weight modules corresponding to the infinite
weight ladder are trivial, and we really only need to consider a finite set
fi pa.,--- , u + qoi. The study of these ladders of weights is important for
the structure theory of modules over semisimple Lie algebras. The direct
sum of the corresponding weight modules, which we may call a weight-ladder
module, can be regarded as a module over the ladder-generating Lie sub-
algebra H L'H Lâ.
2.10 THE FACTORIZATION METHOD AND SPECIAL

FUNCTIONS
The value of the concepts of roots and weights is not limited to semi-
simple Lie algebras. One example of the use of these concepts in a more
general setting arises in the theory of the special functions of mathematical
physics. In addition to the elementary transcendental functions, such as ex
and sinx, these special functions include the Bessel functions, Legendre
polynomials, parabolic cylinder functions, Hermite polynomials and so forth.
Many of them share a variety of interesting properties, such as recurrence
relations, addition theorems, generating functions, orthogonality relations,
integral representations and so on. A partial explanation of these regularities
lies in the fact that many of these functions may be regarded as special cases
of the ordinary or confluent hypergeometric functions. A deeper reason,
perhaps, is that many of these functions arise from the solution of the Laplace
equation upon separating variables. In general, special functions arising from
the solution of the Laplace equation or its generalizations are called harmonic
functions or spherical functions [102], [162], [232]. Now, the Laplace equa-
tion is an example of a differential equation invariant under the action of
a certain Lie group. Hence, it is reasonable to suppose that the theory of
special functions can be unified and also generalized by making use of ideas
2.10 The Factorization Method and Special Functions 71
from the theory of Lie groups and Lie algebras [225] , [234] . An obvious
generalization, for example, would be to replace the Laplace operator for
flat Euclidean space R" by an analogous operator for a curved symmetric
space such as a sphere [243] .
One important method for studying special functions via their recur-
sion relations, known as the factorization method, ties in closely with the
standard Lie algebraic techniques involving roots and weights. Infield and
Hull gave a systematic catalogue of the special functions which can be studied
by this method, which involves factorizing the appropriate second order
differential equation [128]. For the Hermite polynomials, which are related
to the harmonic oscillator, this leads to the creation and annihilation oper-
ators considered previously. Hull and Infield reduced the entire list of
possibilities to six overlapping classes, later shown by Miller to arise naturally
in the representation theory of four particular Lie algebras [168], [169].
These include the Lie algebras of the three-dimensional rotation group and
the groups of Euclidean motions in a plane and in space, as well as a certain
four-dimensional solvable Lie algebra. The Lie algebras that one studies are
those generated by the differential operators which figure in the recursion
relations. The recursion relations for the special functions themselves can
be viewed as examples of the shifting rule for root vectors acting on weight
vectors.
As an example, we discuss the Bessel functions, which are related to the
three-dimensional Lie algebra of the group of Euclidean motions in a plane
[233], [248]. Any Euclidean motion in the plane can be composed out of a
translation and a rotation about the origin. The angle </> of the rotation and
the orthogonal components (x, y) of the translation may serve as local
coordinates for a neighborhood of the identity in the Euclidean group of a
plane. We denote by Tx, Ty and M tangent vectors corresponding to the
one-parameter subgroups consisting, respectively, of translations in the x
direction and y direction, and rotations about the origin. We can identify
the basis vectors Tx, Ty and M for the Lie algebra with differential operators :
This yields an isomorphism of the Lie algebra of the plane Euclidean group
with a Lie algebra of differential operators on the infinitely differentiable
functions on the plane, the Lie product being commutation. Note that the
element
in the enveloping algebra of the Euclidean Lie algebra corresponds to the

Laplacian operator in the (x, y)-plane,
An element of the universal enveloping algebra, such as T2 + T2, which

commutes with the whole Lie algebra is called a Casimir element. In any
module, the eigenspaces of a Casimir element are submodules, and hence
such an element can have only one eigenvalue in an irreducible module.
For our purposes, it will be sufficient to study one such irreducible module,
consisting of eigenvectors of V2 with eigenvalue - 1. A set of solutions for
the corresponding differential equation
is given by the functions
by using polar coordinates (r, <j>) in the (x, ,y)-plane. In fact, it is easy to
transform this equation into the standard differential equation for the Bessel
function Jn(r). The complexification of the real Lie algebra of the plane
Euclidean group is not semisimple since the subspace spanned by Tx and
1y is an Abelian ideal. This Lie algebra is in fact an example of a solvable
Lie algebra which is not nilpotent.
The line spanned by M and the plane spanned by Tx and Ty are both
maximal Abelian subalgebras. However, only the line spanned by M is a
Cartan subalgebra, while the plane spanned by Tx and Ty is not. By setting
the commutation relations take on the simple form
Thus, taking the line spanned by M as our Cartan subalgebra, we may

interpret the three lines spanned respectively by M, T + and T_ as root
spaces. Note that the Laplacian operator factorizes in terms of T+ and T_ ,
Since M^m = imij/m, the vector \j/m is an eigenvector of M, and hence spans
a weight submodule. The elements T + and T_ act as raising and lowering
elements, satisfying the shifting rule
To interpret this formula we use
The shifting rule then reduces to the usual recursion relation for Bessel
functions,
2.11 The Cartan Matrix 73
Many other properties of Bessel functions can be obtained by means of

the Lie algebra ; for example, addition formulas can be obtained by using the
Baker-Hausdorff formulas. Weisner has also obtained generating functions
for certain types of special functions using Lie algebraic ideas [237] .
2.11 THE CARTAN MATRIX
The structure of a semisimple Lie algebra of rank / is determined up to

isomorphism by its system of roots. As a matter of fact, it is even sufficient
to know the system of simple roots a : , , a, with respect to some lexico-
graphic ordering of the roots. The Cartan matrix [A^ defined by
is important in determining the structure of a semisimple Lie algebra.

If a and ft are roots, then the sequence of linear forms
is called an en-ladder through ft if they are all roots and if ft (p + l)a and
ft + (q + l)a are not roots. Then p and q are related by
It follows from the definition of simple roots that the difference a ft

between any two distinct simple roots a and ft cannot be a root. Hence, if
a and ft are simple roots, then p = 0, and the Cartan matrix determines the
a-ladder through the root ft. From a knowledge of the Cartan matrix we
can determine by an algorithm (using induction on the level of the root ft)
whether or not a given linear combination of simple roots ft = i=iMt
is a positive root Thus the full root system is determined by the Cartan
matrix. Actually, at this point it suffices to know that the Cartan matrix
determines the highest root, because the entire root system can be recovered
from a knowledge of the highest root. We later describe a more general
algorithm due to Dynkin which allows one to compute the weight system
of any irreducible module from a knowledge of the highest weight.
The Cartan matrix also determines the Lie products, and hence the
complete structure of a semisimple Lie algebra. Each simple root corresponds
to a certain element
in the Cartan subalgebra. The normalization was chosen to simplify later

computations. The elements h^, , ht form a basis for the chosen Cartan
subalgebra of the semisimple Lie algebra. Let the root vector corresponding
to a simple root a, be denoted by
and let us also define
The vectors e l 5 , et, called simple raising elements, and the vectors
/! , , /, , called simple lowering elements, together generate the whole Lie
algebra. The Lie products are given by
We caution the reader that in these formulas one does not sum over the
repeated indices.
The relations written above do not yet completely define all products,
for (X, ej\ and [/,/,-] have not been specified. The complete set of relations
can nevertheless be derived from the above set by a well-defined algorithm
which we shall shortly illustrate in a special case. The elements hit together
with
and
span the Lie algebra. All linear relations and commutation relations for
these elements are computable from the Cartan matrix.
2.12 THE WEYL GROUP
The roots of a semisimple Lie algebra and the weights of any of its
modules are vectors lying in the dual space H* of a Cartan subalgebra H.
Moreover, these weights, just as the roots, are rational linear combinations of
the simple roots o^, , a,. Hence the weight system of any module over a
semisimple Lie algebra of rank / is a set of points lying in the /-dimensional
real Euclidean space H% of all real linear combinations of the simple roots.
Since this Euclidean space is lexicographically ordered with respect to an
arbitrary basis, we may regard the weights of a module as a totally ordered
set. For a finite-dimensional module, there are only a finite number of dif-
ferent weights, and hence there must be a highest weight which is higher than
all the others. Two irreducible modules over a semisimple Lie algebra are
isomorphic if and only if their highest weights are equal. Thus the irreducible
modules over a semisimple Lie algebra are characterized by their highest
weights.
To study the structure of an irreducible module with a given highest
weight A, we must have a way of calculating its weight system. The weight
2.12 The Weyl Group 75
system has certain symmetry properties which we now study. If fi is a weight

and if a is a nonzero root, then there is a ladder of weights
which also belongs to the weight system. The real number
is an integer, and /i na is one of the weights in this ladder. This transforma-

tion may be interpreted geometrically as a reflection wa in the plane per-
pendicular to the root a in H. These reflections are called Weyl reflections,
and the group generated by them is called the Weyl group of the Lie algebra.
It is clear that the Weyl reflections must satisfy
The weight system of any module is invariant under the Weyl group.
The Killing form also is invariant under the Weyl group. If w is any element
of the Weyl group W, then
In general, the root system may have a higher degree of symmetry than
some of the other weight systems. This is because the root system is not only
invariant under the Weyl group, but also under the inversion a H* a.
The sets of vectors which can represent root systems of semisimple Lie
algebras are restricted by the following three conditions :
1. If a is a root, then a is a root, but no other nonzero multiples of a
are roots.
2. If a and /? are nonzero roots, then
isO, 1, 2 or 3.
3. The Weyl reflection of a root jS in the hyperplane through the origin
perpendicular to any nonzero root a yields another root,
From these three properties of the root system, one can deduce a number
of other simple properties. For example, it follows that the angle between
any two nonzero roots can only be 0, 30, 45, 60, 90, 120, 135, 150 or
180. Also, the relative lengths of any two nonzero and nonorthogonal roots
are determined by the angle between them. If this angle is 0, 60, 120 or 180,
then the roots are of equal length. If the angle is 45 or 135, their relative
lengths are in the ratio ^/2:l, while if the angle is 30 or 150, then this
ratio is >/3:l. In fact, these three conditions alone determine the possible
root systems, and hence all the possible semisimple Lie algebras over the
complex numbers.
2.13 DYNKIN DIAGRAMS
While the problem of finding all semisimple complex Lie algebras has
thus been reduced to a geometrical problem, Dynkin found an even simpler
way of doing this by using diagrams in a plane [81] . He represents a system of
simple roots al , , a, by a set of / points in a plane, where / is the rank of the
Lie algebra. These points form the vertices of the Dynkin diagram. The
points representing a, and a,- are connected by
lines. The number of lines determines the angle between the roots, since
(a,, a,-) ^ 0 for simple roots. Thus the angle 4-a,- / is respectively 90, 120,
135 or 150 if the number of lines which connect the /th and jth vertices is
zero, one, two or three. While the Dynkin diagram of a simple Lie algebra
is connected, the Dynkin diagram of a semisimple Lie algebra which is not
simple is not connected. The connected components of the Dynkin diagram
of a semisimple Lie algebra correspond to the simple ideals occurring in a
direct sum decomposition of the Lie algebra.
The relative lengths of the simple roots are indicated in the Dynkin
diagram by using light and dark circles to represent the vertices. If all roots
are of the same length, all the circles are light. For simple Lie algebras, the
only other possibility is that there are roots of two different lengths. The long
roots are then assigned light circles, the short roots dark circles. The Dynkin
diagram thus allows one to compute the angles between the simple roots and
their relative lengths. The absolute lengths of the roots may be obtained from
an observation of Brown that the sum of the squares of the lengths of all the
roots of a semisimple Lie algebra is equal to its rank [41]. From this informa-
tion one can then draw the root system as a set of points in /-dimensional
Euclidean space. The ratios of the square of the lengths of the simple roots are
related to the Cartan matrix by
Since the product AijAy and the ratio Atj/Ajt can be determined from the
Dynkin diagrams, we can obviously calculate all the Aijt at least up to signs.
The sign is always minus for nonzero off-diagonal elements, and the diagonal
elements are clearly Ati = +2. Thus the Dynkin diagram determines the
Cartan matrix uniquely. From a knowledge of the Dynkin diagram one can
therefore reconstruct the Lie algebra, so that the Dynkin diagram completely
characterizes a given semisimple Lie algebra.
The three conditions which characterize root systems can be reinter-
preted as various restrictions on the possible Dynkin diagrams of simple
2.13 Dynkin Diagrams 77
Lie algebras. For example, a Dynkin diagram cannot contain any loops, and
the total number of lines incident on a given vertex can be at most three.
Continuing in this fashion, we can narrow down the possible connected
Dynkin diagrams to four general classes and five exceptions. Dynkin dia-
grams of the four main classes of simple Lie algebras are shown in Fig. 1,
while those of the five exceptional simple Lie algebras are shown in Fig. 2.
FIG. 1. The four main classes of simple Lie FIG. 2. The five exceptional simple Lie
algebras algebras
These diagrams represent the only possible simple Lie algebras of

finite dimension over the complex number field. Each of these possibilities is
actually realized by a Lie algebra which can be explicitly constructed from
the diagram.
Another useful application of the Dynkin diagram of a semisimple Lie
algebra is found in the description of the structure of the Weyl group. In
fact, the Dynkin diagrams are Coxeter diagrams with additional information
about root lengths. Coxeter initially introduced these diagrams to study
arbitrary reflection groups [67], [68].
The Weyl reflections w, = wa. corresponding to the simple roots a! , ,
a, of a semisimple Lie algebra L, called simple Weyl reflections, are a set of
generators for the Weyl group of L. These generators satisfy the relations
where ptj is 1 if i = j, and is 2, 3, 4 or 6 if i ^ j and the ith and y'th vertices

of the Dynkin diagram are joined by 0, 1, 2 or 3 lines, respectively [47],
[209]. The elements of the Weyl group can be generated on a computer
using a program depending upon this presentation in terms of generators

and relations [24]. This program is based on the observation that S,, the
symmetric group on / letters, is a subgroup of the Weyl group of a simple
Lie algebra of rank /. The computer generates 5, recursively and then gen-
erates the coset representatives of S, in the Weyl group.
2.14 IDENTIFICATION OF SIMPLE LIE ALGEBRAS
The four main classes of Dynkin diagrams correspond to simple Lie

algebras A^B^ Cl and Dt over the complex numbers which are related to
various classical Lie groups. More precisely, we can identify these complex
Lie algebras with the complexifications of real Lie algebras of certain compact
Lie groups. Thus, for example, the complex Lie algebra Al is the complexi-
fication of the real Lie algebra of the compact special unitary group S U(l + 1) .
Similarly, the Lie algebras Bt and Dl correspond to the compact rotation
groups S0(2l + 1, R) and S0(2l, R), respectively, while C, corresponds to the
unitary symplectic group Sp(t). We can also obtain a direct relation between
these complex Lie algebras and the classical groups without complexification
by generalizing the discussion of tangent spaces. Just as we obtain real Lie
algebras as tangent spaces to groups which are real analytic manifolds, we
may obtain complex Lie algebras as tangent spaces to groups which are
complex analytic manifolds. We may then identify Al as the complex Lie
algebra of the complex special linear group SL(l + 1,C) if the latter is
regarded as a complex analytic group. Similarly, the complex Lie algebras
I?, and >, are related, respectively, to the complex orthogonal groups
SO(2l + 1,C) and SO(2/, C), regarding these as complex analytic groups.
Finally, the complex Lie algebras C, correspond to the complex symplectic
group Sp(l, C), leaving invariant a nonsingular antisymmetric bilinear form in
a complex vector space of dimension 21. All of these Lie algebras, which are
defined for all / ^ 1, are simple Lie algebras except for D1 and D2. The
algebra D^ is Abelian, and is thus not simple, while the algebra D2 is semi-
simple, and can be identified with A -i- A-^. For the low-rank Lie algebras
there are a number of isomorphisms,
The dimensions of the main-series complex simple Lie algebras are given by
2.15 COnstruction of the Lie Algebra A2 79
The five exceptional simple Lie algebras G 2 ,F 4 , 6 ,E 7 and 8 over

the complex numbers can be associated in various ways with the Cayley
numbers. These algebras have dimensions 14,52,78,133 and 248, respectively.
The exceptional algebra G2 has been applied by Racah to a problem in
atomic spectroscopy [194].
2.15 CONSTRUCTION OF THE LIE ALGEBRA A2
To illustrate the procedure for determining the structure constants of a

simple Lie algebra from its Dynkin diagram, we consider the simple Lie
algebra A2. The Dynkin diagram for the simple Lie algebra A2 is shown in
Fig. 3.
FIG. 3. Dynkin diagram for A2
The two simple roots o^ and a2 are of equal length, and form an angle of
120. Hence the Cartan matrix, with components Atj, of the Lie algebra A2 is
Since the difference of two simple roots obviously cannot be a root, we have
p = 0 for the a Bladder through <x 2 , and hence,
and this ladder is thus composed of the roots a2 and ct1 + a 2 . By considering
further ladders, we obtain no more positive roots, and we see that o^ + a 2
is the highest root. Thus the root system of the simple Lie algebra A2 consists
of zero, the positive roots a x , a 2 and a t + a 2 and the negative roots
a l s a2 and (a.1 + a 2 ). The Euclidean space H% spanned by the simple
roots <*! and 2 is a plane, and thus we can draw the root diagram of the Lie
algebra A2, as shown in Fig. 4.
FIG. 4. Root diagram for A2

From this figure it is seen that the root system of A2 is not only symmetric
under the Weyl group, but also possesses an additional symmetry under
inversion in the origin.
We obtain part of a canonical basis for A2 by considering the elements
h1,h2 and the simple raising and lowering elements e , e2 and /i ,/2 . Their
commutation relations are
and, of course,
We supplement these six elements with the two further elements
which are root vectors corresponding to the roots (a1 + a2), to obtain a
total of eight vectors which form a basis for A2.
The commutation relations involving the elements el2 and /12 are
obtained by using the Jacobi identity, for example,
In this fashion we obtain the following further commutation relations :
Thus, starting from the Dynkin diagram for A2 we have completely

determined the products of the members of a basis, and hence determined
the structure of the Lie algebra. A similar procedure can be carried through
to construct the Lie algebra corresponding to any Dynkin diagram.
2.16 COMPLEXIFICATION AND REAL FORMS
We recall that there are four general series of simple Lie algebras over
the complex numbers, At,Bt,Ci and D,, and five exceptional simple Lie
algebras, G 2 , F4, E6, 7 and 8. Of course, for / < 4, some of the general
sequence Lie algebras are not defined or they collapse into one another, so
we have only A1, A2, B2, A3, B3 and C 3 . Real simple Lie algebras, such as
2.16 Complexification and Real Forms 81
the algebra so(3, R), are also useful for many applications. The prob-
lem of classifying the real simple Lie algebras is only slightly more
complicated than classifying the complex ones. This is because the process of
complexification relates the real case to the complex case. Hence, we may
study the real simple Lie algebras by generating them from the complex
algebras. Since a single complex simple Lie algebra in general yields several
real ones, we must supplement our study of Dynkin diagrams with further
considerations [93], [94].
We recall that the complexification of a real Lie algebra is obtained by
tensoring it with the complex number field C regarded as a vector space over
the real numbers. Conversely, we obtain the real restriction LR of a complex
Lie algebra L by simply regarding L itself as an algebra over the real subfield
of the complex number field. Note that the processes of complexification and
real restriction are not inverses of each other. The complexification of the
real restriction of a complex Lie algebra Ldoes not yield back L, but rather
C LR L 4- L.
If we can write the real restriction LR of a complex Lie algebra L as a
direct sum L0 i'L0 for some real subalgebra L0, then we call L0 a real
form of L. If L0 is a real form of a complex Lie algebra L, then conversely,
L is isomorphic to the complexification of L0.
Every real simple Lie algebra is either the real restriction or else a real
form of a complex simple Lie algebra. The complication of the classification
problem for real simple Lie algebras arises from the fact that a complex simple
Lie algebra may have several distinct real forms. Or, putting it another way,
several different real simple Lie algebras may have isomorphic complexifica-
tions. Consider for instance the real Lie algebra so(3, R) of the rotation
group and the algebra so(2,1; R) of the plane Lorentz group, that is, the
group of linear transformations on x, y and t leaving x2 + y2 t2 invariant.
These real Lie algebras are not isomorphic, but they have isomorphic
complexifications, namely the complex simple Lie algebra Al.
Of all the real forms of a given complex simple Lie algebra, there is
precisely one, called the compact real form, which is the real Lie algebra of
a compact Lie group. In general, a compact real Lie algebra is defined to be
a real Lie algebra obtained from some compact Lie group. A useful character-
ization of compact semisimple Lie algebras arises from the fact that a real
semisimple Lie algebra is compact if and only if its bilinear Killing form is
negative definite. All noncompact real forms of a simple complex Lie algebra
can be obtained from the compact one [178], [206]. A detailed list of the real
simple Lie algebras, together with their isomorphisms, was found by Cartan
[46], [120].
Any real form L0 of a complex Lie algebra L may of course be regarded
as a subalgebra of the real restriction LR. If et, , en are a basis over the
real numbers for the real form L 0 , then these same vectors are a basis for
L over the complex numbers. Moreover, the vectors ev, , en, ie^, , ien
are linearly independent over the real numbers, and are a basis for the real Lie
algebra L*. We may regard i as a real linear operator on LR having the

properties
for all x and y in LR. Any vector z e L* may be uniquely written as
where x and y are elements of L0. We may define an automorphism 5" of the
real Lie algebra L* by
for all x and y in L0. This is an involutive automorphism; that is, 5 is an

automorphism whose square is the identity operator, S2 = 1. Moreover,
the operator S is the identity when restricted to L0, and its negative when
restricted to the subspace /L0. Thus the real form L0 is the eigenspace of
the linear operator 5 corresponding to the eigenvalue + 1, while /L0 is the
eigenspace corresponding to the eigenvalue 1.
The compact real form of a complex simple Lie algebra L of rank / can
be explicitly constructed as follows. We recall that there is a basis for L con-
sisting of a basis hlt - , ht for a Cartan subalgebra H together with root
vectors ea and e_ a , where a ranges over all the positive roots. It is convenient
to normalize these root vectors so that
The vectors ihlt , ihh together with i(ea + e_J and ea e_x for all
a > 0, then form a basis over the real numbers for the compact real form L0
of the complex simple Lie algebra L. This particular choice of basis makes it
easy to show that the Killing form is negative definite on the compact real
form L0.
Any involutive automorphism 50 of the compact real form L0 of a
complex simple Lie algebra L can be extended to an automorphism of the real
Lie algebra
by defining
for all x and y in L0 . Denoting the eigenspaces of S0 corresponding to the

eigenvalues +1 and 1 by LQ and LQ, respectively, we obtain the Cartan
decomposition
of the compact real form L0 with respect to S0 . We then find that the vectors of
2.17 Real Forms of the Lie Algebra >41 83
are left fixed by S, while those of iL^ are carried to their negatives. To show
that S is an involutive automorphism of
we use the fact that S0 is an automorphism of L0 , so that
for all x and y in L0 . Since L1 consists of all vectors left fixed by S, we see that
if x and y are in Ll , then so is [x, y] , for we have
Hence, Ll = LQ iL$ is a subalgebra of L*, and thus it is a real form of

the complex Lie algebra L. The subspace LQ is called the toroidal part of
the real form L t , while the subspace I'LQ is called its vector part. Since every
real form of L induces an involutive automorphism having the properties we
required of S0 on L0 , every real form of L can be obtained in this manner
from some automorphism S0 of the compact real form L0 .
2.17 REAL FORMS OF THE LIE ALGEBRA A,
The processes of complexification and finding real forms can best be

illustrated by means of a simple example. For this purpose we shall con-
sider the real Lie algebra su(2) w so(3, R) of the ordinary rotation group.
We recall again that this Lie algebra is just a real three-dimensional space
with the usual vector cross product as multiplication. We can take as a basis
a set of elements el, e2 and e3 with the Lie product given by
The complexification of s(2) then has the same basis and same Lie products
over the complex numbers, this complex Lie algebra being, of course, the
complex simple Lie algebra A^. For the benefit of physicists we remark that
it is really this complex Lie algebra Al which is usually studied in the quan-
tum theory of angular momentum. In angular momentum theory it is
conventional not to use the basis elements el,e2,e3, but to replace them
with an obviously equivalent set
To discuss the real forms of the complex Lie algebra Alt we use still
another basis, which corresponds to the root space decomposition of this
algebra. The Cartan subalgebras of the complex simple Lie algebra At are
one-dimensional. The selection of a Cartan subalgebra is described in
angular momentum theory as choosing a quantization axis. It is conventional
to choose h = ;3 = ie3 as the basis of a Cartan subalgebra, but of course
one could equally well have chosen jl or j2 for instance. Note however that
there are one-dimensional subalgebras, such as the one spanned by the

vector J! + i/2 > which are not Cartan subalgebras. With the chosen vector
h as a basis for the Cartan subalgebra H, the dual basis for H* is given by the
linear form a on H satisfying
The roots of the Lie algebra A x with respect to H are the three linear forms 0,
+ a and a. The corresponding root vectors are
and
which together with h form a basis for the three-dimensional complex Lie
algebra A1 . We have
The compact real form of A^ is the original real Lie algebra su(2) *
so(3, R), since the corresponding Lie groups SU(2) and SO(3, R) are both
compact. The compact real form can also be obtained from the general
prescription given earlier. According to (his prescription, we should have
as a basis for su(2), which it obviously is.

To obtain the other real forms of Ait we study the involutive auto-
morphisms of the compact real form. Since an involutive automorphism
S0 of sw(2) satisfies S% = 1, it can only have 1 as eigenvalues. If all its
eigenvalues were +1, then S0 would be the identity, and the real form
generated by it would just be su(2) again. The condition that S0 be a Lie
algebra automorphism is
for all x and y in the Lie algebra If all the eigenvalues of S0 were - 1, then
S0 = 1, and
while
so that S0 could not be an automorphism. Thus, any nontrivial involutive

automorphism must have at least one eigenvalue + 1 and one eigenvalue - 1,
and the eigenspaces LQ and LQ are both nonzero. Letting x and yt be cor-
responding eigenvectors, we have
so that [x,y] is also an eigenvector corresponding to the eigenvalue 1,

that is,
2.17 Real Forms of the Lie Algebra A, 85
Similarly the products
and
follow from the fact that S0 is an automorphism. Since [x,y] is the vector
cross product of x and y, it is a nonzero vector orthogonal to both x and y,
and hence LO must be two-dimensional, containing the two linearly in-
dependent vectors 3; and [x, y], while LQ is one-dimensional. The vectors
3; and [x,y] are an orthogonal basis for the eigenspace LQ, and since
[LQ,LQ] c LQ, we see that
But this implies that [y, [x, y]] is a multiple of x because LQ is one-dimensional,
and hence x is orthogonal also to y. Thus x, y and [x, y] are mutually orthog-
onal vectors, and if x and y are taken to be unit vectors, then [x, y] will
also be a unit vector. We can assume that et = x, e2 = y and e3 = [x, y]
without loss of generality. There is thus, up to isomorphism, only one
noncompact real form of A t , corresponding to the automorphism
of the compact real form L0 = su(2). Since LQ is spanned by et, and LQ is

spanned by e2 and e$,a basis for the noncompact real form Ll = LQ iL$
is given by the vectors e^, ie2 and ie3. The Lie products of these basis elements
are of course given by
The real Lie algebra so obtained is the real Lie algebra of several Lie
groups, including the plane Lorentz group SO(2,1; R) leaving invariant
the form
the real special linear group SL(2, R), the real symplectic group Sp(l, R)
and the pseudo-unitary group 517(1,1). All of these Lie groups are, of course,
locally isomorphic since their Lie algebras are isomorphic. The representa-
tions of these noncompact groups have been studied extensively [17], [32],
[193].
The real Lie algebras related to the complex Lie algebra Al include not
only the two real forms discussed above, but also the real restriction A* of A.
The real restriction A* is a six-dimensional real Lie algebra, a basis being

given by
The real Lie algebra A* is the Lie algebra of several familiar Lie groups,
which are again locally isomorphic. These include the ordinary homogeneous
Lorentz group 50(3,1; R) leaving invariant the form
the complex special linear group SL(2, C) and the complex orthogonal group
50(3, C).
2.18 ANGULAR MOMENTUM THEORY
It may be of interest to further elaborate on the use of complexification

and real forms in angular momentum theory. The theory of the rotation
group is, after all, one of the main applications of Lie algebraic ideas to
quantum mechanics. This theory, also known as the quantum theory of
angular momentum, is extensively used in atomic and nuclear physics.
The complex Hilbert space of all quantum mechanical state vectors
may be regarded as a module over the complex Lie algebra A of the three-
dimensional rotation group. Thus the elements of Al act as linear operators
on this Hilbert space. An explicit realization of these operators can be
obtained by quantizing the classical theory. The quantum mechanical
operator corresponding to the classical orbital angular momentum
for a single particle is obtained by replacing the momentum p by the dif-

ferential operator ih\, obtaining
The operators L are Hermitian, and they satisfy the commutation

relations
where we have set h 1 for convenience. Adding also the spin S, for which
there is no completely satisfactory classical analogue, we obtain the total
angular momentum
for a single particle. One now postulates that the operators J are also
Hermitian, and satisfy the same commutation relations as L, namely,
2.18 Angular Momentum Theory 87
The formal theory of angular momentum is built up on the basis of

these commutation relations and Hermiticity requirements alone [84],
[203]. One of the main roles of the Hermiticity assumption in angular
momentum theory is just to argue that the unitary representations of the
rotation group must be finite-dimensional. If we drop this Hermiticity
requirement, then the resulting theory can also be applied to the finite-
dimensional nonunitary representations of the Lorentz group. To discuss
the representation theory of A one customarily defines operators
so that the commutation relations become
If \l/ is an eigenvector in Hilbert space such that J3\j/ = mi/f, then J\j/
are either zero or else eigenvectors of J3 with eigenvalues m 1. This
shows that J+ raises the eigenvalue by one unit and J_ lowers the eigenvalue
by one unit ; hence we have the terms "raising" and "lowering operators."
The theory of the Lorentz group is the relativistic generalization of
angular momentum theory. To develop this theory one may proceed
heuristically, starting with the classical description of a relativistic spinning
particle. The equation
generalizes to the formula
for the generators of the homogeneous Lorentz group, where /x, v = 0, 1, 2, 3.

Here X = t is the time and P = E is the total energy, and we have set the
velocity of light equal to unity, c =* 1. The energy and momentum of course
must be given by their relativistic expressions. The momentum of a free
particle with rest mass m and velocity u is given by
while the energy is given by
Since M MV is antisymmetric, it has six independent nonzero components,

including the three components of angular momentum
and three components called Lorentz boosts,

The components of J generate ordinary rotations, while those of K generate

pure Lorentz transformations, that is, Lorentz transformations which
preserve a space-like plane.
For the corresponding quantum mechanical theory, these classical
quantities must all be replaced by Hermitian operators defined on a common
dense domain of Hilbert space and satisfying appropriate commutation
relations. One is naturally led to postulate the commutation relations
where the metric tensor g"v is + 1 if n = v = 1, 2 or 3, and g = 1, while

all the other components vanish. Since these commutation relations imply
that the position operators X" fail to commute with the mass operator
P^Pp, they are seldom used in practice for the description of particles having
a definite mass [13]. Instead, the kinematics of relativistic particles with spin
is completely described in terms of the generators Af MV of the Lorentz group
and the momenta Pv. Together, the six generators M MV of the Lorentz group
and the four components Pv of momentum span the ten-dimensional Lie
algebra of the Poincare group. Practical applications of the Poincare
group in elementary particle physics deal with the helicity representation
for partial- wave expansions of high-energy scattering amplitudes [132],
[172].
The Poincare group, or inhomogeneous Lorentz group, is the group of
transformations of the four-dimensional Minkowski space-time which
preserve distances between points. The Poincare group is not semisimple
since the set of all parallel translations of space-time forms an Abelian normal
subgroup. The homogeneous Lorentz group is defined to be the quotient
group of the Poincare group modulo its translation subgroup. The Lorentz
group may alternatively be described as the set of all elements of the Poincare
group which leave fixed an arbitrary specified point of space-time. The whole
Poincare group is then generated by the translation subgroup and any one of
these Lorentz subgroups. The corresponding breakup of the Poincare Lie
algebra as a direct sum of the Abelian ideal generating translations and a
simple subalgebra generating Lorentz transformations is a Levi decomposi-
tion for this algebra.
To clarify the structure of the Poincare and Lorentz Lie algebras further,
we examine the commutation relations for Afv and Pv. The relations
show that the translation subalgebra is Abelian, while
shows that it is an ideal. The generators of the Lorentz subgroup satisfy the
commutation relations
2.18 Angular Momentum Theory 89
Although the real Lie algebra X? of the Lorentz group is simple, its
complexification is not simple, but only semisimple, splitting up as A r + Al.
We may note, by the way, that S0(4, R) and S0(3, R) x SO(3, R) have real
Lie algebras isomorphic to sw(2) 4- sw(2), whose complexification also
yields Al + Al. In other words, both Xf and sw(2) + su(2) are real forms of
the same semisimple complex Lie algebra A^ + Al.
As a consequence, the theory of the Lie algebra of the Lorentz group
may be described as a doubled version of angular momentum theory. In
fact, the commutation relations of the Lorentz algebra, which may of course
be rewritten in terms of J and K, acquire a simple form if we set
What emerges then is the explicit At + Al structure of the complexified Lie

algebra of the Lorentz group,
Although J and K are Hermitian operators, the operators J* are not,

and because of this, the Lorentz group can have infinite-dimensional unitary
irreducible representations [182].
CHAPTER 3
CONSTRUCTIVE METHODS
3.1 RAISING AND LOWERING SUBALGEBRAS
Any module over a Lie algebra may also be regarded as a module over
its universal enveloping associative algebra. We may therefore use the
representation theory of associative algebras to clarify the structure of Lie
modules. For example, the universal enveloping algebra plays an important
part in constructing the irreducible modules over a semisimple Lie algebra.
For this purpose we need to take a quick look at the structure of the envelop-
ing algebra of a semisimple Lie algebra L resulting from its root space de-
composition. We choose a Cartan subalgebra H and denote the subspaces
of L spanned respectively by the root vectors of the positive and negative
roots by L+ and L~'. The subspaces L+ and L~ are nilpotent subalgebras of
L generated respectively by the simple raising and lowering elements, and
we can write
Corresponding to this breakup of the Lie algebra L into H, L+ and

L~, we can also break up the universal enveloping algebra U(L) into three
parts U, U+ and U~. Here 17 is the Abelian subalgebra generated by the
identity element and the Cartan subalgebra H, while U+ and U~ are gener-
ated by the identity element together with L+ and L~, respectively. We
may call U+ and U~ respectively the raising algebra and lowering algebra
of the Lie algebra L. The whole enveloping algebra U(L) may be written as
the product of the three subalgebras U, U* and U~, taken in any order.
The algebra U+ is generated by the identity and the simple raising elements,
while [/"is generated by the identity and the simple lowering elements. Since
the Cartan subalgebra H is spanned by the commutators of simple raising
and lowering elements, the whole universal enveloping algebra U(L) is
generated by the identity and the simple raising and lowering elements.
The lowering algebra U ~ is a useful tool for obtaining the direct sum
decomposition of a reducible module M over a semisimple Lie algebra L.
An extreme vector x in a module M is a nonzero vector which is annihilated
by all the simple raising elements el, , el in the sense that
91
92 Constructive Methods 3.1
for i = 1, , / . Since the root vector ea of any positive root a is generated by

the simple raising elements el, , et, any extreme vector x also satisfies
the more general condition
for a > 0. The extreme vectors together with the zero vector form a subspace,
the extreme subspace of the module. If a module is irreducible, its extreme
subspace is one-dimensional, and is spanned by a weight vector v for the
highest weight L Indeed, for any a > 0, the vector ejo would be a weight
vector for the higher weight A + a if it were not zero. In any module M, the
subspace 17 ~x is an irreducible submodule if x is an extreme vector. More-
over, any basis X j , , xn of its extreme subspace yields a decomposition
of the module M as a direct sum of irreducible submodules
Thus a knowledge of the extreme subspace yields a reduction of the module

to a direct sum of irreducible submodules, which is a valuable technique, for
instance, in computing Clebsch-Gordan coefficients.
Irreducible modules can be generated not only by their extreme vectors,
but by any nonzero element. To do this, however, we have to use the whole
enveloping algebra, not just the lowering subalgebra. Thus if x is any element
of an irreducible module M over a Lie algebra L, then we have M = U(L)x.
That is, every vector yeM can be written as y = ux for some element
u e U(L). By use of this result it becomes possible to construct the irreducible
modules themselves out of the enveloping algebra U(L). To do this, we need
not assume that the Lie algebra L is semisimple, and we may treat a class of
modules slightly more general than the irreducible ones.
A module M over a Lie algebra L is called a cyclic module if there is a
vector x in M such that
If we regard U(L) as a module over itself, the mapping which takes any
element u E U(L) into the vector ux e M may be described as a module
homomorphism from U(L) onto M. The kernel of this module homomor-
phism is a submodule, and hence a left ideal, of U(L) which we call the anni-
hilator of the cyclic module M. The annihilator A does not depend on x, and
may be described as the set of elements a e U(L) such that av = 0 for all u e M.
By the fundamental homomorphism theorem, the module M is isomorphic to
V(L)/A, the element MX in M corresponding to the coset u + A in U(L)/A.
This result, M w U(L)/A, means that every cyclic module may be obtained
as a quotient of U(L) with respect to some left ideal. The problem of finding
all the irreducible representations of U(L), and consequently of L, is equiva-
lent to that of finding all the maximal left ideals of U(L). Conversely, if A
is a left ideal of U(L), then the cosets form a cyclic module U(L)/A over L,
and this module is irreducible if and only if A is a maximal left ideal.
3.2 Dynkin Indices 93
3.2 DYNKIN INDICES
The problem of classifying the irreducible modules over a semisimple

Lie algebra has been reduced to finding all possible highest weights. We may
use the lowering algebra to show that the set of vectors in H which can
be highest weights for irreducible modules is closed under addition. To see
this, let Ml and M2 be irreducible modules with extreme vectors x and x2
corresponding to the highest weights A x and A 2 , respectively. Then xx x2
is an extreme vector in the tensor product module M: M 2 , and applying
the lowering algebra, we obtain the irreducible submodule U~(xl x2)
having A! + A2 as its highest weight. The submodule l/~(x! x2), sometimes
denoted by M M 2 , is called the Carton composition of the modules
A/\ and M 2 . The Cartan composition of two irreducible modules is that
irreducible submodule of their tensor product which corresponds to the
highest weight. A basic weight is a highest weight which is not the sum of two
nonzero highest weights. The number of basic weights of a semisimple Lie
algebra is equal to its rank /. An irreducible module is said to be a basic
module if its highest weight is one of the basic weights.
The basic weights satisfy a number of interesting relations which make
them useful as a basis for the space H . We can number the basic weights
AI , , A( to correspond to the simple roots a1 , , a, so that
is zero if i ^ j and one if i = j. From these formulas and a knowledge of the

simple roots, one can construct the basic weights either geometrically or
algebraically. Any weight ^ is a linear combination of basic weights,
with integer coefficients
called the Dynkin indices of the weight //. The Dynkin indices of the highest
weight uniquely characterize an irreducible module over a semisimple Lie
algebra up to isomorphism. In general the Dynkin indices of a weight can be
any integers, positive, zero or negative, but the Dynkin indices of a highest
weight are always nonnegative. The Dynkin indices of the simple roots a,-
are given by the elements of the Cartan matrix [Atj] , where
and we may thus write

The action of the Weyl group on the weight system of a module is particularly
easy to compute in terms of Dynkin indices. Each basic weight is invariant
under all but one of the simple Weyl reflections since w^- = kj for i ^ j,
while w^ = A, <X; . Hence the Dynkin indices m- of the reflected weight
Wj/2 are related to the indices ml of /i itself by the relation
We may construct an irreducible module with highest weight
in terms of a set of basic modules Mlt , M,. This module is the Cartan
composition
where the weight vector x,- in M,-, for the basic highest weight A,-, appears nt
times in the tensor product. Thus the process of Cartan composition reduces
the problem of constructing the irreducible modules of a semisimple Lie
algebra to that of constructing its basic modules.
Dynkin introduced a convenient way of specifying the irreducible
modules over a semisimple Lie algebra [83]. One simply writes the rth
Dynkin index of the highest weight above the ith vertex of the Dynkin
diagram of the Lie algebra. In Fig. 5 this is illustrated for a particular rep-
resentation of the Lie algebra B3, the complexification of the real Lie algebra
ofSO(7,R).
FIG. 5. Dynkin diagram for the ^-dimensional spinor module ofB3, which can be used to define
the exceptional Lie algebra G2
This particular module is the 8-dimensional spinor module of B3,

which is of special interest because it can be used to define the exceptional
simple Lie algebra G2 One may in fact obtain G2 as the subalgebra of all
elements in B3 which annihilate a certain element of this module [27].
Besides the Dynkin diagram notation, we shall also use the "common
notation {N} to denote an irreducible module of dimension N. The main
advantage of this notation is that it is concise, while the Dynkin diagram
notation has the advantage of supplying more pertinent information which
is useful for computations. In some cases, there are several nonisomorphic
irreducible modules with the same dimension, which may be distinguished by
adding primes, asterisks, and so forth. For example, the dual of the module
{N} is denoted by {N*}. For Lie algebras of type B, or Ct and for Lie algebras
of type DI with even /, as well as for the particular algebras At, G2, <F4, E7
3.3 Irreducible Representations of A, 95
and 8, each module is isomorphic to its own dual [161]. The dual operation
thus plays no particularly important role for these Lie algebras, while for
the remaining algebras, many but not all modules are self-dual. In practice,
the main application of duality in the representation theory of simple Lie
algebras is to Lie algebras of type Al with / ^ 2. For an irreducible module
over A{, the Dynkin indices of the dual module may be obtained by replacing
m,- b y m / + 1 _ .
3.3 IRREDUCIBLE REPRESENTATIONS OF A,
We now discuss an important example, the computation of the ir-

reducible representations of the complexified Lie algebra A^ of the ordinary
rotation group SO(3, R). Any representation of S0(3, R) is a homomorphism
into the group of nonsingular linear operators on a vector space, and its
kernel is a closed normal subgroup. The kernel must be all of S0(3, R) or
just the identity, and such a representation either trivially maps all of 50(3, R)
onto the identity, or it is a one-to-one mapping. The composition of any
representation of S0(3, R) with the canonical projection of its covering
group SU(2) onto S0(3, R) yields a representation of the group SU(2).
Moreover, the image of any normal subgroup of SU(2) under the canonical
projection is a normal subgroup of SO(3, R), and hence the only proper
closed normal subgroup of SU(2) consists of +/. The representations of
SU(2) which can be obtained from those of 5O(3, R) are those which are not
faithful, having kernels which contain the negative of the identity, /.
When we pass to the simple complex Lie algebra A, we obtain all the
representations of SU(2), not just those which arise from the representations
of the rotation group. Instead of the usual basis j1 J2,J3 for the Lie algebra
A1} we again use the basis consisting of h = j3 and ea = j = y'j ij2,
which satisfies
There is a Cartan subalgebra H with basis h, and roots 0, <x, where <x(/z) = 1.
Since the sum of the squares of the lengths of the roots equals the rank, which
is one, we have (a, a) = |. The raising and lowering algebras U+ and U~
consist, respectively, of all polynomials in ea and in e_ a . There is one basic
weight ^. in the one-dimensional space H%, and since it is a multiple of a
satisfying
it must be 1 = a/2.
In the quantum theory of angular momentum, the basic module over
A! is known as the spinor representation, or the spin-half representation of
the rotation group [190]. The weight vector x corresponding to the highest
weight A satisfies
and is an extreme vector, so that
Since the basic weight A goes over into its negative under a Weyl reflection,
the weight system of the basic module consists at least of +A, and in fact
consists of these two weights alone. Since
it follows that y e.gX is nonzero and hence is a weight vector for the
shifted weight A - a = -A. We must have e.j = 0, because if it were
nonzero, we would obtain a weight lower than A, and by Weyl reflection
one higher than A. Hence the spinor space U~x is a two-dimensional complex
vector space spanned by the two basis vectors x and y. The action of A j on the
basic module is given by
Rewriting this action in terms of the basis JiJ2>J3 yields the Pauli
matrices,
The linear mapping which takes jk into %ak for k = 1,2,3 is precisely what is
called the spin-half representation of the Lie algebra Av.
Every irreducible module over Av is isomorphic to a module generated
by letting the lowering algebra U~ act on a tensor product x (g> x.
Each irreducible module is therefore characterized by a single integer n,
the rank of the tensor x x, that is, the number of repetitions of x.
The tensor x x is the highest weight of the module, and since
the highest weight for the module is (w/2)a = ya. In the terminology of
angular momentum theory, the number j = n/2, which can be either an
integer or a half-integer, is called the spin of the module [84]. The irreducible
modules corresponding to representations of the rotation group are just
those having integer spin, while the others are representations only of its
covering group. We now calculate the entire module structure, choosing
3.4 The Casimir Subalgebra 97
the module basis to agree with the usual conventions of the quantum theory
of angular momentum. In the case n = 2, corresponding to spin one, we
find
For symmetry, it is convenient to take as basis for this module
More generally, for the irreducible module M, with spin j, we can compute
a basis of 2j + 1 vectors um, where
with
(2j factors),
This basis can be chosen so that the action of the Lie algebra is given by
These formulas are central in the theories of the representations of the

ordinary rotation group and of the homogeneous Lorentz group [84],
[182].
3.4 THE CASIMIR SUBALGEBRA
The center C(L) of the universal enveloping algebra U(L) is a key

element in the computational aspects of representation theory [147],
[229]. It is an Abelian subalgebra consisting of those elements z in U(L) which
commute with every element of the Lie algebra L. For a semisimple Lie
algebra L, there is always one such element which can be constructed as
follows. Let elt , en be any basis for the Lie algebra L; since L is semisimple,
its Killing form is nonsingular, and there exists another basis el, , ? for
L such that
The element
of the universal enveloping algebra, called the second order Casimir element
of L, commutes with every element of the Lie algebra [49]. The second
order Casimir element for a semisimple Lie algebra is analogous to the
Laplace operator in the theory of special functions. We can also give a
formula for y which is highly reminiscent of the factorization method for
the Laplace operator. If h1, , ht is a basis for a Cartan subalgebra H,
then since H is nonsingular, there is another basis h1, , h1 such that
The second order Casimir element can then be expressed as
The sum in the second term goes over all the nonzero roots a of L, and ea is
any nonzero root vector in L.
The second order Casimir element is used in the proof that modules
over semisimple Lie algebras are completely reducible [50]. It also has many
other uses in the theory of representations of semisimple Lie algebras. In an
irreducible module M, the operator corresponding to y is a multiple of the
unit operator, y = y(M) 1. If M has highest weight A, then the eigenvalue
y(M) of the operator representing the second order Casimir invariant in M
is given by Weyl's formula
where <5 is half the sum of the positive roots. In the adjoint representation of
a simple Lie algebra L, we have y(L) = 1, and thus the operator correspond-
ing to y is the unit operator itself. For example, if we take the usual angular
momentum basis jt, j2, ;3 for the Lie algebra Av, then the Casimir element y
is
which has the eigenvalue
in an irreducible module with spin j. For the adjoint representation of A,,

the spin is unity, j 1, and jj(j + 1) = 1. The second order Casimir element
is a simple example of an element of C(L), and we may think of C(L) as
consisting of the Casimir element and its various higher order generalizations.
Since the subalgebra C(L) of the universal enveloping algebra contains what
are known to physicists as higher order Casimir invariants, it is natural to
call C(L) the Casimir algebra of L. A Casimir element is any element in the
center C(L) of the universal enveloping algebra.
3.5 Irreducible Representations of A2 99
For irreducible representations each element of C(L) is represented by

a multiple of the identity. Recently, much work has been done to find the
relation between the eigenvalues of these higher order Casimir elements and
the classification of irreducible modules by their highest weight [5].
For any finite-dimensional module M, we can define a linear form T on
the Casimir algebra by means of the trace,
where ze C(L). The linear form T e C(L)* characterizes the modules in the
sense that if modules M and M' have linear forms T and T' which are pro-
portional to each other, then the modules are isomorphic [116].
3.5 IRREDUCIBLE REPRESENTATIONS OF A 2
The simple Lie algebra A2 is the complexification of the real Lie algebra
of the special unitary group SU(3). This particular group has recently been
studied extensively in applications dealing with the strongly interacting
elementary particles [72]. We recall that the six nonzero roots of A2 have
equal lengths, forming the vertices of a regular hexagon, as shown in Fig. 4
( 2.15). Since the sum of the squares of the lengths of all the roots is equal to
the rank, which is two, each nonzero root has length 1/^/3. The Killing form
can now be computed since the angle between the simple roots a t and a 2 is
120. We find that
and
The canonical basis for A2 consists of the simple raising elements e and
e2 , the simple lowering elements /t and /2 , their commutators el2 = [elt e2]
and /12 = [/i,/2] and the elements hv and /i2. The equations defining the
basic weights ^ and A2 can be solved either algebraically or geometrically.
These equations say that Aj is perpendicular to a 2 , and its projection onto
ocj is half the length of a x , as shown in Fig. 6.
FIG. 6. Basic weights of the simple Lie algebra A2 shown superimposed on the root diagram
Every representation of A2 can be obtained from the two basic ones.

The representation which corresponds to the basic weight A2 is just the dual
of the one corresponding to A t . In the applied literature dealing with the
elementary particles, these are known as the quark and antiquark representa-
tions.
FIG. 7. Weight diagrams of the two basic modules ofA2
Using the Weyl group, we find that the weight system of the basic module
with highest weight A t is an equilateral triangle consisting of three weights
/d , fi2 and {JL3 , as shown in Fig. 7. For the basic module Ml we choose a basis
*i , x2 , x3 consisting of weight vectors corresponding to the weights H i , n 2 , f * 3
respectively, where
Having chosen the extreme vector x t , clearly we can define xz and x3 by the
equations
fixing their relative normalizations.

We can now use the commutation relations of the Lie algebra A2 to
compute the matrices for the representation. As a sample calculation, we
note that
where we used
and A//JJ-) = 6^. The shifting rule LM c M^+a and an inspection of the
weight system shows that
so that the only nonzero entry of the matrix corresponding to ^ is in the

first row and second column. A more complete calculation leads to the
3.6 Characters 101
following matrix representation corresponding to the basic module

over the Lie algebra A2:
The weight system of the dual module M2 M* is obtained from that

of M! by inversion in the origin. Thus if u^, u2 and u3 are the weights of the
basic module Mlt then / x l f -u2 and u3 are the weights of the basic
module M 2 . If we use the dual basis x*, x*, x* in the dual module M*, then
the representation matrices in the dual module are the negative transposes
of those computed above.
The representation matrices in an arbitrary irreducible representation
of A2 can be obtained by a calculation similar to the one performed above.
By using the second and third order Casimir elements of A2, one can also
derive general formulas for these matrices [14].
3.6 CHARACTERS
The character of a finite-dimensional module M over a group G is a

complex- valued function % : G - C defined for all g in G by
Here TrMg denotes the trace of g, where we regard g as a linear operator on

the vector space M. Isomorphic modules have the same character because
the trace is invariant under a similarity transformation.
To make the transition from group theory to Lie algebra theory in the
discussion of characters, we proceed as follows. Let G be a compact and
connected Lie group corresponding to a semisimple complex Lie algebra,
obtained for example by exponentiation. The subgroup corresponding to a
Cartan subalgebra is a maximal torus subgroup, that is, a compact Abelian
subgroup isomorphic to the direct product of a number of circle groups.
Every element of the group G is conjugate to some element of this maximal
torus subgroup [73], [92]. We recall that elements gl and g2 are conjugate
if there is an element s in the group such that g2 = sgjS" 1 . Since the character
takes on the same value at conjugate elements of the group, it is determined
by its restriction to such a maximal torus subgroup.
This leads us to define the character of a finite-dimensional module
over a semisimple complex Lie algebra as a complex-valued function on a
Cartan subalgebra. The character % of a module M over a Lie algebra L with
Cartan subalgebra H is defined by
for all h in H. In this definition of the character we regard h as a linear operator

acting on the finite-dimensional vector space M, and the exponential function
is given by its convergent Taylor series.
There are a number of elementary properties of characters. For example,
setting h = 0, we get the trace of the unit operator in the module M, which
is equal to its dimension:
Characters are also useful for computing the Clebsch-Gordan series which
reduce the tensor products of irreducible modules as direct sums of ir-
reducible submodules. The character of the direct sum M + N of two
modules M and N is the sum of their characters :
This may be proved by choosing a basis in the direct sum which is the union
of bases in the summands. The character of the tensor product M N of
two modules is the product of their characters :
This may be proved by substituting
into
and interchanging the order of summation.

The representation matrices with respect to the dual basis in the dual
of a given module over a Lie algebra are obtained from those in the original
module by taking the negative transpose. Since the trace is unaffected by the
operation of taking transposes, we see that
3.7 Computation of the Killing Form 103
Since any module M is the direct sum of its weight submodules M,

we can write
We define the multiplicity n^ of a weight \i to be the dimension of the corre-

sponding weight submodule :
Consequently the character 7 of a module M may be written as
where the summation goes over all the weights of H in M. This formula
shows that a knowledge of the character of a module is equivalent to knowing
its weight system and the multiplicity of each weight. Practical methods for
computing characters and multiplicities of weights will be discussed later.
We may note that the highest weight of an irreducible module over a semi-
simple Lie algebra has multiplicity one. Also, if w is any element of the Weyl
group, then the weights fj. and wî have the same multiplicity in any module.
3.7 COMPUTATION OF THE KILLING FORM
The first step in performing calculations on the representations of a

semisimple Lie algebra on an electronic computing machine is the computa-
tion of various quantities related to the Lie algebra itself. This phase includes
setting up the Killing metric of the Lie algebra as well as its Cartan matrix and
the inverse of the Cartan matrix [83]. The determination of these matrices
depends on the choice of simple root labeling. Our choice for the labeling of
the simple roots may be seen by inspecting the Dynkin diagrams depicted
in Fig. 4 ( 2.15). Also, the Weyl group and the system of positive roots need
to be determined for some of the later calculations. These algorithms can be
carried out efficiently using integer-mode arithmetic on a computer [4],
[26]. The avoidance of floating-point arithmetic is desirable not only because
from theoretical considerations every number is an exact integer, but more
importantly because later computations involve tests for equality of
numbers which could be seriously affected if roundoff errors were permitted.
Roots and weights can be represented as integer arrays by using Dynkin
indices, which are their components with respect to a basis consisting of the
basic weights A u , /I,. In particular, we recall that the entry Atj of the
Cartan matrix is the ith component of the y'th simple root a,. The Killing
metric is a rational matrix [g,-_,-], where
We can write
where [Gy] is an integer matrix and D is a common denominator. Then we

can compute the inner product of any weights n = m^ and v = ,- nfa
since
A simple way to compute gtj is to start from the quantities
which can be read off directly from the Dynkin diagram. Since the Dynkin
diagram informs us directly only about relative lengths and angles between
simple roots, we only obtain Mtj and we will later have to compute the
normalization factor N by separate means. We can compute the integer
Cartan matrix from
and its inverse may either be obtained directly by a modified Gauss-Jordan

reduction procedure or by consulting tables given by Dynkin [83] . Overflow
problems can be minimized in the inversion routine by using small nonzero
integers as pivot elements and repeatedly reducing fractions to lowest terms,
these methods being sufficient up to rank 8. The inverse [A^ *] is a rational
matrix which is represented in the machine as a ratio
The formula
(no sum) yields the metric up to the normalization factor D which we can
obtain in two different ways. One way, which requires a prior computation
of the root system, is to use the fact that the sum of the squares of the lengths
of the roots of a semisimple Lie algebra is equal to its rank. For simple Lie
algebras, a better procedure is to use the fact that the second order Casimir
operator has the value 1 in the adjoint representation. If we know the highest
weight of the adjoint representation, that is, the highest root, then this
method is computationally much simpler. Indeed, if rj , , r, are the Dynkin
indices of the highest root, then
and
3.7 Computation of the Killing Form 105
This is just the Weyl formula
where p is the highest root and <5 is half the sum of the positive roots.
We may illustrate the method for the simple Lie algebra B2 correspond-
ing to the Lie groups SO(5, R) and Sp(2). From the Dynkin diagram shown
in Fig. 8, we obtain
by inspection, and the Cartan matrix is thus
The inverse of the Cartan matrix for B2 is
and the metric is given up to normalization by
Since the Dynkin indices of the highest root of B2 are (2, 0), we find D = 24
and N = 6, and hence the matrix [g0-] is given by
FIG. 8. Dynkin and Root diagrams of B2, showing also the basic weights
3.8 DYNKIN'S ALGORITHM FOR THE WEIGHT SYSTEM
The basic inputs for a given weight diagram calculation are the rank
and type of the simple Lie algebra under consideration and the Dynkin
indices of the highest weight of the irreducible module whose weight diagram
is to be determined. The outputs include the weight system, the multiplicity
of each weight, and other pertinent information. Given the highest weight
of an irreducible module over a low rank semisimple Lie algebra, finding all
the other weights is a simple problem in geometry. Since the Weyl group
describes the symmetry properties of weight systems, we need only apply
Weyl reflections to the highest weight to obtain the general shape of the
weight system. We can then fill in the rest of the weight system by using
ladders of weights.
To illustrate this procedure, consider again the simple Lie algebra B2 .
The Weyl group of this algebra is generated by the Weyl reflections w t and
w2 associated with the simple roots a x and a 2 . If we introduce Cartesian
coordinates (x, y) in the plane of the root system, then w^ and w2 are the
following maps :
reflection in the x-axis,
reflection about the line
Note that
where 1 is the identity mapping, since WiW 2 is a 90 rotation. The Weyl

group, consisting of the elements 1, w^, vv 2 , w 1 w 2 , \v2w1, W 1 w 2 w 1 , w 2 w 1 w 2
and (WiWj) 2 , is a non-Abelian group of order eight which may be identified
as the dihedral group of a square [69] . The two basic modules over B2 have
dimensions four and five, respectively. The weight systems of these basic
modules, shown in Fig. 9, are easily obtained by applying Weyl reflections
to their highest weights and filling in ladders.
For higher rank semisimple Lie algebras, the geometrical method
becomes impractical because we cannot easily draw figures in many dimen-
sions. What we want then is a simple algorithm for finding the weight system
from the highest weight which can be used in a program for an electronic
computer. Dynkin gave one such algorithm, in which the weight system is
FIG. 9. Weight diagrams for the basic modules of B2

3.8 Dynkin's Algorithm for the Weight System 107
split up into layers, the first layer consisting of the highest weight alone [83],
[135]. Each weight in a given layer may be obtained from some parent
weight n contained in the previous layer by subtracting some simple root a,.
To obtain a particular layer, given all the preceding layers, we consider the
arladders passing through all the weights
belonging to the previous layer. Since the top and bottom weights of the
ladder
are related by the simple Weyl reflection wt corresponding to the root a,,
we have
Now WZ/Y = n - m-a, and wâ, = -a,, so that we find
Since the weights lying above // in this ladder belong to previous layers, the
value of q for this ladder is known and we can compute p = q + m . The
weight /j. a,- belongs to the given layer if and only if p is positive.
We illustrate this algorithm in Fig. 10, which summarizes the computa-
tion of the weight system for the 16-dimensional irreducible module over
B2 with highest weight (1, 1).
FIG. 10. Computation of weights for the 16-dimensional irreducible module for B2 with highest
weight
From the transpose of the Cartan matrix we can read off the simple
roots, which turn out to be (2, -1) and (-2,2). The weight system, con-
sisting of a total of 12 weights, is shown in Fig. 11.
FIG. 11. Weight diagram (showing multiplicities) for the irreducible module for B2 with highest
weight A A, + <12
In the process of determining the weight system, it is a gopd idea to

order them lexicographically as well. For this, we do not use the Dynkin
indices directly, but first calculate the components of a given weight
fi = ]T m,-A,- with respect to a basis consisting of the simple roots. We do so
because lexicographic ordering is defined in terms of the components of a
weight with respect to the simple roots rather than in terms of the Dynkin
indices, which are their components with respect to the basic weights. We
find that
where
and B is the matrix related to the inverse of the Cartan matrix. Then /* > 0,
for example, if and only if the first nonzero entry in the new array (p 4 , , p{)
is positive. Thus the process of lexicographic ordering requires a previous
computation of the inverse of the Cartan matrix. The final output of the
Dynkin layer method is a table of the Dynkin indices of all the weights of a
given irreducible module, listed in lexicographic order.
The root system may be computed by the same procedure used to
determine the weight systems. To do this, of course, it is necessary to know
the highest root of the Lie algebra, which is listed in Fig. 12.
3.9 Freudenthal's Algorithm 109
FlG. 12. Dynkin indices of the highest roots
For several algorithms to be described later, it is necessary to know all

the positive roots. To obtain these, we may apply the Dynkin algorithm to
the adjoint representation to obtain the roots, and then lexicographically
order the roots to find the positive ones. Another method of calculating the
positive roots is to use explicit models of the Lie algebras, in which case we
do not need to lexicographically order the roots [81].
3.9 FREUDENTHAL'S ALGORITHM
To compute the character X A of an irreducible module M with highest

weight A over a semisimple Lie algebra L, we need a method for finding the
multiplicities of all its weights. There are a number of methods available for
computing characters of Lie modules which can be applied uniformly to
any type of simple Lie algebra. One of these methods dates back to the early
researches of Weyl in 1925-1926 on the representations of semisimple Lie

groups [239], [240], [241], [242]. The Weyl method makes use of the Weyl
group to generate an object called a girdle, and the character is obtained as
a ratio of two girdles. A second method, due to Freudenthal, makes use of
a simple algorithm for computing the multiplicities of the weights [89] , [90] ,
[91] , [92] . This method has the advantage that it avoids the group generated
by Weyl reflections, and can therefore handle Lie algebras of large rank. A
third method, due to Kostant, also makes use of the Weyl group, but differs
from the Weyl method in that the girdle division is replaced by partition
function methods [145], [146], [226]. Racah's formula (also developed by
Klimyk [140], [196]), involving the Weyl group but no partition function,
provides a recursive expression for the multiplicity of the weights. Each of
these algorithms presupposes that the weight system has previously been
determined. Kostant's method as well as Weyl's original method are some-
times considered to be more elegant than Freudenthal's or Racah's method
from a theoretical point of view in that these methods make use of closed
formulas rather than recursive procedures.
The real question however is which of these algorithms is most efficient
when implemented on a computing machine. To study this question,
we have recently experimented with various implementations of these
algorithms in FORTRAN and ALGOL. These computations were carried out
on the UNI VAC 1108 computer at Carnegie-Mellon University and on the
IBM 360/75 system at the University of Pennsylvania [4], [25]. All of these
algorithms are practical procedures for computing characters of modules
over Lie algebras of rank |4. The Weyl group rapidly becomes very large
with increasing rank, and therefore takes up considerable computing time
as well as causing storage problems. If it is only necessary to compute the
characters of modules with relatively low dimension, it would seem to be
desirable to avoid computing the Weyl group altogether. With the Freuden-
thal algorithm we were still able to compute characters of modules of dimen-
sion up to 1000 for Lie algebras of rank ^ 8. Thus, this method can be used
for all five exceptional simple Lie algebras G 2 , F4, E6, 7 and 8- However,
the last three exceptional simple Lie algebras E6, 7 and E8 have compara-
tively few representations of dimension less than a thousand. On the other
hand, for low rank and high dimension the Kostant formula appears to be
best, especially if the multiplicities of only certain weights are to be computed.
The Freudenthal algorithm is based on a formula which may be used
as a recursion relation to find the multiplicity nM of a given weight ^ from
the multiplicities of higher weights. This recursion relation is
where 6 - \ J>0 a denotes half the sum of the positive roots. The sum over
a goes over all positive roots, which are known from the root system. The
sum over j is not really infinite but breaks off after a finite number of terms.
3.10 The Weyl Character Formula 111
To use this formula to compute the multiplicities, we start with the multi-
plicity of the highest weight A, which we know is unity for an irreducible
module. If ^ is a weight other than the highest weight, and if we know the
multiplicities for all weights // > //, then each term on the right side of the
preceding expression is known. Since for any weight /j. ^ A, the quantity
is nonzero and positive, we can always solve for n^ . The use of the Freudenthal
formula can be a lengthy process, but it can be done automatically on a
computing machine [4], [26]. Sometimes Freudenthal's algorithm may be
made more efficient by making limited use of Weyl reflections and by using
the fact that weights related by Weyl reflections have the same multiplicity
[148].
As an example of the use of Freudenthal's formula, we consider the
simple Lie algebra B2 . We have already computed the metric :
The positive roots are a l 5 a 2 , a: + a 2 and 2a.1 + <x 2 , and half their sum is
5 = A! + A 2 . Now consider again the 16-dimensional module over B2 with
highest weight A = Aj + A2 which we previously discussed. Its weight
system consists of an octagon with a square inside it. Since the eight weights
of the octagon are related to the highest weight by the Weyl group, these
must have the same multiplicity as the highest weight, that is, one. The four
weights of the square inside the octagon are also related to each other by
Weyl reflections, and we thus need only compute the multiplicity of one of
these. Applying the Freudenthal formula to the case we calculate
In the summation over j appearing in the Freudenthal formula only the

term with j = 1 is nonzero, and we obtain nu = 2. Thus the four weights
forming the square inside the octagon all have multiplicity two, and the
dimension of the module is 8 x 1 + 4 x 2 = 1 6.
3.10 THE WEYL CHARACTER FORMULA
The Weyl formula for computing characters is conceptionally somewhat

simpler than the Freudenthal algorithm. However, Weyl's formula does not
turn out to be any easier to implement on an electronic computer. We discuss
the Weyl formula mainly because it gives, as a corollary, a useful formula for
calculating the dimensions of the irreducible modules, by using the fact that
The Weyl formula can be proved by purely algebraic means, and has recently
also been extended to algebraic groups [48], [218].
For a semisimple Lie algebra, Weyl's formula expresses the character

Xt(h) of the irreducible module having the highest weight A as a ratio of
two girdles, each girdle also being a linear combination of e^'s. The coeffi-
cients of the eM(/l)'s in a girdle are 1, so that the girdles are simpler than
the characters themselves. The main computational problems in using the
Weyl formula are the generation of the Weyl group needed to compute the
girdles and the division of the one girdle by the other. For low rank Lie
algebras, the Weyl group is easily obtained, but the girdle division is extremely
tedious to carry out by hand. Examples of such calculations for simple Lie
algebras of rank two have been given by Behrends, Dreitlein, Fronsdal and
Lee [27]. For rank 3, some of these computations are also available in the
literature [143], [154].
The coefficients 1 occurring in the formula for a girdle are the parities
of elements of the Weyl group. The Weyl group Wis composed of reflections
and products of reflections, each reflection being an orthogonal linear
operator with determinant equal to 1. The parity det w of an element w
in the Weyl group W is equal to + 1 if w is a product of an even number
of reflections and 1 if w is a product of an odd number of reflections.
We define the k-girdle as
where 6 is half the sum of the positive roots. This quantity 5 is also equal
to the sum of the basic weights :
The Weyl character formula expresses the character of an irreducible

module with highest weight A as the ratio of two girdles :
The 0-girdle appearing in the denominator is also essential in the derivation

of Freudenthal's algorithm.
To clarify the Weyl formula, we introduce some further terminology
regarding girdles. The idea here is to work with the algebra of polynomials
in the quantities e*** and allow the Weyl group to operate directly on this
algebra. These quantities form a group under multiplication, and the char-
acters and girdles are elements of the group algebra of this group. The
group algebra of an infinite group G over the complex field is the set of all
formal sums g6G cgg, where the cg are complex numbers and cg = 0 for all
but a finite number of elements g e G. Sums and scalar multiples of elements
in the group algebra are defined componentwise, while multiplication is
defined by
3.10 The Weyl Character Formula 113
The group algebra is infinite-dimensional with G as a basis, but no con-

vergence questions arise in these definitions since each element has only a
finite number of nonzero components.
The weights // of any module belong to the set J of all integer linear
combinations of the basic weights /lj , , A, for a semisimple Lie algebra L.
The additive group J is clearly the direct sum of / infinite cyclic groups
generated by the basic weights. If e* denotes the function which assigns
the complex number e"('l) to the element h e H, then we may write the char-
acter as x ZMS./ nftfl- Here we have formally extended the sum so that //
varies over J, this being possible because we have n^ = 0 when \L is not a
weight. The multiplicative group of the quantities e* is isomorphic to J since
The algebra of polynomials in these quantities may thus be identified with

the group algebra of J. The identity element of our algebra is e, and it is
clear that the algebra is both commutative and associative. We turn this
group algebra into a module over the Weyl group as follows. For any
element w of the Weyl group W, we define the action of w on an element of
the group algebra by
We define an antisymmetrizer
which may be regarded as a linear operator on the group algebra. The

definition of the A-girdle may then be written as
where 6 is half the sum of the positive roots, and the Weyl formula may be
written as
The character ^ of a finite-dimensional irreducible Lie module with

highest weight A is sometimes called a primitive character. We recall that if
two modules are isomorphic, then their characters are equal. To establish
the converse, we write
Since the elements e* form a basis for the group algebra, if the characters
of two irreducible modules are equal, then their highest weights are equal,
and hence the modules are isomorphic. Any finite-dimensional module M
over a semisimple Lie algebra is completely reducible. This means that M
can be written in one or more ways as a direct sum of irreducible modules.

If m ; is the number of irreducible modules with highest weight A in one such
decomposition, then the character % of M is given by
Since the primitive characters are linearly independent, the m A are uniquely
determined, and to this extent at least the reduction of M is unique.
3.11 THE WEYL DIMENSION FORMULA
To obtain the dimension of the irreducible module MA from the Weyl

character formula, we have to set h = 0. The Weyl formula gives the char-
acter as a quotient of two girdles
We cannot simply set h = 0 in this expression since 0(0) = 0, but we must

instead evaluate the limit of this expression as h - 0, using
Since H% is a Euclidean space, we can regard H as a complexified Euclidean

space, and then questions of analysis, such as taking limits, are well-defined.
We can thus use L'Hospital's rule to compute the limit as h - 0 to
obtain dim M. First, one can show that the 0-girdle can be written in product
form as
We thus see that, for small /i,
To obtain the limit as h - 0, we let h = thd, where t is a real number and

h6 is an element of H such that
This procedure allows us to relate the /l-girdle to the 0-girdle since one can
show that
Thus we can obtain ;,(thd) for small t from the corresponding formula for
0(th8) simply by replacing S by A + <5, giving
3.11 The Weyl Dimension Formula 115
This yields the Weyl formula for the dimension of the irreducible module
with highest weight X :
In using this formula on a computer, there is some danger of overflow

for high rank algebras because of the many positive roots. This problem may
be minimized by reducing fractions to their lowest terms at frequent intervals
in the computation. Better still, one may use to advantage the representation
of any nonzero fraction as
where a, /?, y, are integers of either sign or zero.

As an example of the Weyl dimension formula, we consider any irre-
ducible module over the simple Lie algebra A t . The root system for A l is
a, 0, a so that there is only one positive root, and thus, half the sum of the
positive roots is
The space H| is one-dimensional and is spanned by the basic weight
The highest weight of an irreducible module over A^ may be written as

A = nAj , where n is a nonnegative integer, the Dynkin index of the module.
Recall that j = n/2 is an integer or half-integer called the spin and
A = na./2 = jet. We compute the dimension of any irreducible module with
spin j using the Weyl dimension formula. Since there is only one positive
root, we have only one factor, and
dimA^ = (a, A + <5)/(a,(5) = 2j + 1.

For the algebra Alt it is practical to use the Weyl character formula
itself. The Weyl group for the algebra A l is
and the formula for the girdle reduces to two terms. Also, wa/^ = /i, and
hence,
The character of M A is given by the Weyl formula as

Thus, the weights are
each with multiplicity 1.
3.12 CHARACTERS OF MODULES OVER THE ALGEBRA A2
As a less trivial application of the Weyl character formula, we consider

the simple Lie algebra A2, related to the compact Lie group SU(3). Let A
be the weight system of the irreducible module over the simple Lie algebra
A2 having highest weight A, and let
Thus \}/i is defined almost exactly like the character &, except that n^ is
replaced by unity for all weights fi belonging to the weight system A. Thus
if/1 can be calculated from a knowledge of the weight system alone, without
knowledge of the multiplicities of the weights.
The weight system in general has a hexagonal-type shape, as shown in
Fig. 13, and may be obtained from the highest weight A = pAj 4- qX2 by
the process of Weyl reflection and filling in ladders.
The girdles of the simple Lie algebra A2 satisfy the recursion relation
from which we may obtain by iteration a formula for the character [7] . For
triangular weight systems (those with either p or q equal to zero) we have
while for nontriangular weight systems, the recursion relation yields the
formula
FIG. 13. Weight diagram of the 105-dimensional representation D(6, 2) of A2, showing the
multiplicity of each weight
3.13 The Kostant and Racah Character Formulas 117
Here A is a multiple of either Xv or A2 and n is a positive integer, while

<5 = A! + A2 as usual is half the sum of the positive roots. If p ^ q, we may
let A = (p ^H! and n = q, so that X + nd = /?A, + <?A2. For p < q
we simply interchange the roles of p and q, setting A = (4 p)A2 and n = p.
These considerations lead to the following prescriptions for obtaining
the character of any irreducible module with highest weight A = p^ + <?A2
over the algebra X 2 . It suffices to consider the case p ^ q, because inter-
changing p and q is equivalent to taking the dual module. The weight system
of the dual module can be obtained by inversion of the weight system of the
original module with respect to the origin. This amounts to turning the
figure upside down.
For p ^ q the construction is as follows. First draw a hexagonal-type
figure, with sides of alternating lengths of p and q units, each unit being the
length of the nonzero roots, which is 1/^/3, so that the horizontal side at
the top has length p/^/3, for example. The interior angle between adjacent
sides is 120. On every side whose length is p (respectively 4) units, we then
place p + 1 (respectively q + 1) points spaced at unit intervals, as in Fig. 13.
If q = 0, the figure degenerates into a triangle. The interior of this figure is
now filled with further points arranged in an equilateral triangle-type array,
the distance between any two adjacent points being one unit.
This gives us the weight system and it remains to specify their multi-
plicities. The weight system may be regarded as a series of hexagons of
decreasing size, one contained in the other, eventually degenerating into
triangles. All weights on the outer perimeter of the figure have multiplicity
one. Proceeding inward, each successive hexagon consists of weights whose
multiplicity is one more than the previous hexagon, until the multiplicity
reaches the value q + 1, at which point there is no further increase of multi-
plicity. In other words, just at the point that the hexagons degenerate into
triangles, the multiplicities of the weights stop increasing. Explicit drawings
of weight diagrams of modules over A2 can be found in the literature [72],
[106].
3.13 THE KOSTANT AND RACAH CHARACTER FORMULAS
A number of detailed hand computations of characters were published

recently in connection with speculations about hadron symmetry schemes
in elementary particle physics [183], [217]. We have already mentioned the
explicit formula for the multiplicity of any weight of an irreducible module
over A2 which Antoine discovered. Similar, but rather more complicated
formulas can be obtained for the other two simple Lie algebras of rank two.
The case of the Lie algebra B2, corresponding to the Lie groups S0(5, R)
and Sp(2), has been considered by Gruber and Weber, and by McConnell,
while the case of G 2 has been discussed by Radhakrishnan and Santhanam,
and by McConnell [111], [165], [166], [199]. For rank three, there are three
simple Lie algebras A3 % D3, B3 and C3 over the complex numbers. Some
early calculations of weight diagrams for the rank three simple Lie algebra
A3, corresponding to the Lie groups 5(7(4) and SO(6, R), were given in a
paper by Wigner over thirty years ago [247]. Some examples of weight
diagrams for rank three simple Lie algebras have been given by Loupias,
Sirugue and Trotin [154]. The most complete hand calculations for rank
three were carried out by Konuma, Shima and Wada [143]. For higher rank
algebras, fewer results have been reported in the literature. Examples of
weight multiplicity tables for the exceptional simple Lie algebra F4 are given
by Dynkin and Veldkamp [83], [231] . Speiser has given the weight diagrams
for the basic modules of Lie algebras of the types At, Bl and >r(see [216]).
Several other multiplicity tables have also appeared in the literature [90],
[218] . The difficulty of the hand computations led to considerable discussion
of improvements in computational methods [109], [110].
Kostant gave another character formula which is closely related to the
Weyl character formula. The Kostant formula provides a closed form
expression for the multiplicity n^ for any weight fj. of the irreducible module
with highest weight A over a semisimple Lie algebra [145], [146]. In a sense,
the Kostant formula is obtained from the Weyl character formula by
formally carrying out the girdle division. To do this, one uses a partition
function P(/i) defined for all linear forms // in H% as the number of ways of
writing /z as a linear combination of positive roots with nonnegative integers
as coefficients. That is, P(fi) is the number of ways of writing
where ka ^ 0 is an integer and a a positive root. Kostant's formula is
It was suggested by Antoine and Speiser that the Kostant formula

might save computation time, particularly for high rank Lie algebras and
for high-dimensional representations [8] , [9] . The computation of the parti-
tion function is a standard problem of combinatorial analysis [158], [199].
We have P(0) = 1 since there is only one way of writing zero as a nonnegative
integer linear combination of positive roots. Setting A = 0 into the Kostant
formula, we obtain the following recursion relation, valid for /i ^ 0 :
This recursion relation can be used as a method for computing the

partition function. Tarski used this method to compute by hand the partition
function for the algebras A2, B2, G2 and A3, but the formulas became very
involved [226]. A computer implementation of the Kostant formula has
been reported [25].
3.14 The Steinberg and Racah Formulas 119
Racah [196] suggested another formula for characters and this method
has also been described by Klimyk [140]. The Racah formula provides a
recursion relation directly for the multiplicities of the weights:
It does not involve the partition function. A computer implementation of

this formula has also been reported [25].
3.14 THE STEINBERG AND RACAH FORMULAS FOR

CLEBSCH-GORDAN SERIES
One of the main applications of characters is to reduce tensor products

of irreducible modules over a semisimple Lie algebra, that is, to calculate
the Clebsch-Gordan series. Suppose we want to reduce the tensor product
Mj (x) M 2 of irreducible modules Ml and M2 having highest weights ^
and A 2 respectively. The outer multiplicity m*^^ of a linear form X e J is the
number of irreducible submodules with highest weight X occurring in the
Clebsch-Gordan series for the tensor product MlM2. These outer
multiplicities can be computed from the characters by the formula
Given the weight diagrams, this formula could be used directly to compute
the outer multiplicities by a process of repeated subtraction. The leading
term on the right is the one for
and by examining the coefficient of e*1 + *2, we see that
Subtracting this term from both sides and examining the leading terms of
the remaining expression yields the outer multiplicity of the next highest
submodule, and so on. This procedure can easily be programmed for a
computing machine, and thus each method for computing weight diagrams
yields a corresponding method for computing Clebsch-Gordan series. In
practice, this procedure for computing the outer multiplicities can be made
more efficient [140], [155]. For example, if the Weyl girdle method is chosen,
only one girdle division is needed if we rewrite the above formula as
Steinberg used the Kostant formula to develop a method for calculating

these Clebsch-Gordan series directly [219]. The Steinberg formula for
m*^ involves a double summation over the Weyl group:
Some applications of the Steinberg formula have been given in the literature
[198], [221], [222].
The Steinberg formula is a generalization of the original Clebsch-
Gordan formula for computing the tensor product of modules over the
simple Lie algebra A 1 . The character of the irreducible module M^ with
spin j, highest weight A = jet, and dimension 2j + 1 is
We may use this character formula directly to reduce the tensor product
M A (g) Mr of two such modules as a direct sum of irreducible submodules.
Setting /I = ja and A' = /a, we find that the character of the tensor product
module is
The Clebsch-Gordan series for A are thus given by
a familiar result in angular momentum theory [203]. These results follow

immediately from the Steinberg formula. Since the Weyl group of Al has
two elements, 1 and w, the Steinberg formula has four terms only. We may
set A = jet, A! = â and A2 = y'2a, and note that 3 = a/2 and wp = n for
any weight JJL. Thus the Steinberg formula reduces to
For AI, the partition function P(n) is unity if /* = na, where n is an integer
^0, and zero otherwise. Clearly m A Al ^ 2 = 0 unless jl + j2 j is an integer.
Since 7, ^ 0, j2 ^ 0 and 0 ^ ; ^ ^ + j2 , the last term in the Steinberg
formula is always zero. Of the two terms with negative signs, only one can
contribute at a time, and it does so when \jl - j2\ > j, canceling out the
contribution from the first term on the right. Thus we find that m*^^ = 1
when y'i + j2 - j is an integer and l^ - j2\ ^j^ Ji + J2, while m\ lA2 = 0
otherwise.
3.14 The Steinberg and Racah Formulas 121
We shall also illustrate the use of the Steinberg formula for the simple
Lie algebra A2, corresponding to the group S U(3). The Cartan matrix of A2 is
and its inverse is
Hence the basic weights Aj and A2 are related to the simple roots o^ and a 2 by
From the root system, shown in Fig. 4 ( 2.15), we see that cil , a 2 and a t + a2
are the positive roots.
Using the Weyl formula, we find that the dimension of an irreducible
module with highest weight plv + qk2 is given by
In particular,
and
so that we may write
We now show by means of the Steinberg formula that
or in other words,
To apply the Steinberg formula, of course, we need to use the Weyl group.
The Weyl group W of the simple Lie algebra A2 consists of the elements 1,
w
i > W 2> W i w 2> vv 2 W! and WIH^W, = w 2 WiW 2 . This group is isomorphic to
the symmetric group S3 of all permutations on three elements, the corre-
spondence being given by
The action of the Weyl group on the positive roots is shown in Fig. 14
FIG. 14. Action of Weyl group on positive roots
By looking at dimensions, the only candidates for submodules in the

reduction of {3} {3*} are {1} and {8}, and possibly also the modules
{6} = M2^ and {6*} = M2i2. Using the Steinberg formula, we find that
the multiplicity of the module with highest weight 0 for example is
Similarly, the multiplicity of the module with highest weight /l x + A 2 is

also 1, so that indeed
These results could also have been obtained by other means, besides the
Steinberg formula. For this Lie algebra, and more generally also for the
simple Lie algebras of types A-D, there are some frequently used alternative
methods involving Young tableaux for computing Clebsch-Gordan series
[115].
Another formula for computing Clebsch-Gordan series is due to
Racah. It is a recursive formula which avoids the partition function and the
double summation over the Weyl group, but does require the computation of
the weight diagram of the module M^ with highest weight A t . Racah' s
formula is
where ni+s_w(iz+S) is the multiplicity of the weight A + d w(A2 + <5) in

the module M j with highest weight A- t .
3.15 TENSOR ANALYSIS
We shall now turn to the explicit construction of modules over simple

Lie algebras, using tensor analysis. The classical methods of tensor analysis,
suitably modernized, are useful for such practical matters as computing
Clebsch-Gordan and Racah coefficients, which are much used in applica-
tions. The degree of difficulty of such computations frequently depends on the
3.15 Tensor Analysis 123
availability of suitable methods to construct the irreducible modules, and

some flexibility in this respect is often welcome. For this reason, the knowl-
edge of several different ways to construct a particular irreducible module is
not a useless luxury. The classical procedure is to try to generate all the
irreducible modules from some particular one. By taking tensor powers of
this particular module, and applying such processes as symmetrization and
contraction, one may hope to construct all the other" irreducible modules.
This leads us to study the general classification of tensors according to their
symmetry properties and degree of tracelessness.
The construction of all modules has been pushed back from general
modules to irreducible modules by means of direct sum decompositions and
from irreducible modules to basic modules by means of the Cartan composi-
tion. By the use of tensor analysis, we can continue this process back one
more step, arriving finally at the concept of an elementary module [83], [216].
An elementary module is a basic module corresponding to a terminal node of
the Dynkin diagram. It is a module from which we can generate the other
basic modules by tensorial methods. The elementary modules are thus the
natural building blocks out of which to construct all possible modules.
We shall begin our discussion of tensor analysis by considering Lie
algebras of type Al because this is the simplest case. The basic modules for
these algebras are specified by their highest weights as shown in the Dynkin
diagrams in Fig. 15.
Module Highest weight Dynkin diagram
FIG. 15. Basic modules for the simple Lie algebra A
The simple Lie algebra Al may be identified with the linear Lie algebra
consisting of all traceless linear operators in an (/ + l)-dimensional vector
space M over the complex number field. The vector space M may itself be
regarded as an irreducible module with highest weight ^, and may thus be
identified with the basic module M j . The module M is an elementary
module over the simple Lie algebra At, and the other basic modules M 2 ,
M 3 , , M, can be constructed as exterior powers of M. Explicitly, the
basic module Mk is isomorphic to the fcth exterior power of the elementary

module:
(k copies).
One could equally well construct all modules starting with the dual module
M * M,, so that the dual module M* may also be regarded as an elementary
module.
Similar procedures can be applied, with varying degrees of success,
for the simple Lie algebras 5,, C, and Dt. For these algebras, a natural module
to use as a building block would be the one whose Dynkin diagram is
listed in Fig. 16.
Algebra Dynkin diagram of

elementary module
FIG. 16. Elementary modules for tensor analysis in the simple Lie algebras Bt,Ct, Dt
For the orthogonal algebras B, and >,, the exterior powers A * M again yield
all but one or two of the basic modules. The omitted modules are the basic
spinor module of B, and the two basic semispinor modules of D{, which we
shall discuss later. For the symplectic Lie algebras C,, the exterior powers
A * M are not irreducible, and to obtain the basic modules this method must
be supplemented by a trace removal process.
3.16 YOUNG TABLEAUX
Once the basic modules have been constructed from the elementary
ones, we can then construct all irreducible modules via Cartan composition,
making use of the enveloping algebra. In this way, in principle, the entire
representation theory is reduced to a single elementary module. For some
applications it is advantageous to bypass the Cartan composition procedure
and construct the irreducible modules directly from the elementary module
via tensor analysis. To do this, one needs to study the symmetrization of the
tensor powers of the elementary module M. We consider here the case of
algebras of type Al for which this procedure is fairly straightforward [115],
[131]. The tensor power module (X)"M may be considered not only as a
module over the Lie algebra At, but also as a module over the permutation
group Sn. For each permutation n in Sn, there is a linear operator on the
3.16 Young Tableaux 125
tensor power module (X)"M which permutes the various factors in the
tensor product. We define the action of n on the tensor product of the vectors
xl5 , xninMby
The idea now is to construct projection operators by using the group

algebra of the permutation group, making use of the theory of Young
tableaux [204]. (The latter will be defined below.) For example, if dK denotes
the parity factor 1 of the permutation n, we can define an antisymmetrizer
in the group algebra by
The exterior power A " M is then naturally isomorphic to the antisymmetric

submodule j/ (X)" M of the tensor power.
To go further one uses some results from the Wedderburn theory of
semisimple associative algebras. In addition to the antisymmetrizer, we
can construct other operators which project irreducible components out
of the tensor powers of the elementary module. There is in fact a basis ^
for the group algebra of the permutation group Sn satisfying
For a given a, the elements e^ generate a simple ideal of the associative

group algebra of the permutation group. The elements e* e'r are seen
to be projection operators, (e?)2 = e*, and give a resolution of the unity
element of the group algebra :
The ideals of the group algebra of the permutation group can be labeled
by means of Young shapes as follows. A partition [a] of an integer n is a
list of integers [a x , , ak] such that al ^ a 2 ^ ^ a k > 0 and
The partition [a] can be denoted by a Young shape consisting of n boxes

arranged so that a, boxes occur in the jih row. If there are several a's which
are equal, then we may use an abbreviated notation, where am means that
a is repeated m times. For example, the partition [22 1] is represented by
the Young shape
A Young tableau is a Young shape in which the boxes have been num-
bered from 1, , n in some order. A standard Young tableau is a tableau
with the property that if all but the boxes labeled 1, , h with h < n are
erased, then these remaining h boxes form a Young tableau by themselves.
An example of a standard Young tableau is
The number fa of standard Young tableaux for a given Young shape [a]
where x{ = a,. + k - i. The standard Young tableaux corresponding to a

given shape [a] can be labeled [a]r, where r runs from 1 to /a. The particular
method of numbering the standard tableaux is a matter of convention.
The theory of Young tableaux leads to an explicit construction for the
basis elements e^ of the group algebra of the permutation group Sn. To carry
out this construction one defines, for each Young tableau [a]r, a pair of
subgroups called the positive and negative groups. The positive group P[a]r
contains all permutations which permute only the numbers within each row,
while the negative group N[a]r contains all permutations which permute only
numbers within each column of the tableau. For example, the permutation
(13) (25) belongs to the positive group of the standard tableau given in our
example, while (124) belongs to the negative group. We define a symmetrizer
P?, which is the sum of all the elements of the positive group, and an antisym-
metrizer N", equal to the sum of all elements of the negative group, each
element being multiplied by its parity sign factor. Explicitly, the symmetrizer
on all the elements in rows of the tableau is given by
and the antisymmetrizer on the columns of the tableau is given by
Let a"s denote the permutation which transforms the tableau [a]s into
the tableau [<x]r :
3.16 Young Tableaux 127
and define
The elements Ears form a basis for the group algebra of the permutation group,
but do not have simple multiplication laws. To obtain simpler basis ele-
ments, we first expand E*s in terms of all the permutations n in Sn :
and we denote by *s the coefficient of the identity permutation e in this

expansion
Let (a )~ l = rja denote the matrix inverse to the matrix whose elements are
the coefficients gs. Then the elements
form a basis of the group algebra of the permutation group, and these have
the simple multiplication properties mentioned earlier. It turns out that
the matrix rf is in fact very close to being a unit matrix, and for the elements
with r s, with which we are primarily concerned here, these coefficients
effectively disappear, and we have simply
Thus the projection operators can be computed directly from the positive
and negative groups of the tableau, without requiring a calculation of the
>tfs coefficients [204].
The permutations n in Sn commute with the elements x in the Lie
algebra and hence so do the elements e*> from which it follows immediately
that e*Mn is a submodule of the module M". The modules e*M" for the
same Young shape a, but different tableaux r and s, are isomorphic :
For algebras of type A{ , these modules are all irreducible, and each irreducible
module can be obtained in this way by projecting out symmetrized portions
of the tensor powers of the elementary module [244] . In general, however,
there will be infinitely many different Young shapes corresponding to a given
irreducible module. To obtain a one-to-one correspondence between
isomorphism classes of irreducible modules and Young shapes for Lie
algebras of type At, we may restrict the class of Young shapes under con-
sideration to those with at most / rows. The irreducible modules can then be
characterized either by such a restricted Young shape [a] or by the highest
weight L For a Young shape [a] = [o^ , , aj with a total of n boxes, the
highest weight of the modules e?M" is
where ^ , , A, are the basic weights and we set a,+ x = 0. Since the pro-
jection operators ef form a resolution of the identity, we also have
giving a complete reduction of the tensor powers of the elementary module

into its irreducible components. This is equivalent to the classical reduction
of a tensor with an arbitrary number of indices to a sum of tensors with
definite symmetry properties.
3.17 CONTRACTIONS
For Lie algebras of types Bh Ct and Dt> the reduction of the tensor
powers of a module M into symmetrized tensor powers e*Mn is not complete
because these submodules need not be irreducible [6]. To remedy this, we
need to discuss the process of contraction of tensors and the removal of
traces [195].
For the module M we shall take the space traditionally used to define
the algebras Bt, C{ and />,. It has 21 + 1 dimensions for the odd-type orthog-
onal Lie algebras Bt, and it has 21 dimensions for both the even-type orthog-
onal Lie algebras D, and the symplectic Lie algebras Ct. The defining
module M is equipped with a nonsingular bilinear form (x,y) which is
symmetric in the case of the orthogonal Lie algebras and antisymmetric
for the symplectic Lie algebras. In both cases, the Lie algebra elements a e L
are linear operators on M which are antisymmetric with respect to the
bilinear form
For i, j = 1, , n we define the contraction operator
by
A circumflex is used here to indicate that the corresponding factor is to be

omitted. The contraction operators are easily seen to commute with all
elements of the Lie algebra. The induced bilinear form on M", defined by
3.17 Contractions 129
allows us to talk about orthogonality for tensors. One can now classify
tensors according to their degree of tracelessness, and thus arrive at the
orthogonal direct sum decomposition
Here M" consists of completely traceless tensors, that is, elements of M"
which are annihilated by any contraction. The space M"2 consists of tensors
which are orthogonal to all traceless tensors but which are annihilated by
a product of any two contractions and so forth, producing spaces Ml which
are mutually orthogonal and linearly independent. Finally, the remainder,
if any, is the subspace M tt0 orthogonal to all of the subspaces Ml with k
= 1, 2, , [n/2], which therefore consists of those tensors most resistant
against being annihilated by contractions.
As an example of such a decomposition into traceless tensors we may
consider an orthogonal Lie algebra. In the module M we may introduce an
orthonormal basis ev , , en satisfying
The tensor product module M may then be decomposed as
where M I is spanned by the single tensor "= l et e^ and M2 is spanned

by the tensors e{ (x) e^ and ef et e-t ejt where i ^ j.
The submodules Ml of the tensor power M" are not yet irreducible in
general, but since they are modules over the permutation group Sn as well
as over the Lie algebra, we may carry out a further reduction by symmetriza-
tion. The submodules e?M are finally irreducible modules over the Lie
algebra, but they are not all distinct, and there are various isomorphisms
between these modules [115]. In addition, for the orthogonal-type Lie
algebras there exist modules which can only be obtained by supplementing
the methods of tensor analysis with the methods of spinor analysis.
For the symplectic Lie algebras C /5 the exterior powers A " M of the
defining module M fail to be irreducible, and the contraction process enters
into the construction of the basic modules. As a typical example, we shall
study the case of the simple Lie algebra C3 . The Lie algebra C3 is related to
the symplectic Lie group Sp(3), which may be considered as a group of
6 x 6 complex matrices, and therefore the 6-dimensional basic irreducible
module appears to be a natural starting point for building up the others.
The Dynkin diagrams of this module and of the other two basic modules, as
well as of some other modules, are shown in Fig. 17.
FIG. 17. Dynkin diagrams for some of the irreducible modules for the Lie algebra C3
The Dynkin diagrams for the complete weight system of the fundamental
6-dimensional module and the simple roots of C3 are given in Fig. 18.
The weight system here was calculated by using the Dynkin algorithm
discussed earlier (3.8).
The following Clebsch-Gordan series of modules over C3 can be
obtained, for instance, from a study of characters :
Simple roots Weight system for the fundamental

6-dimensional module
FIG. 18. Roots and fundamental weights for the simple Lie algebra C3
3.18 Spinor Analysis and Clifford Algebras 131
The Dynkin diagrams for the various modules which appear here are
among those included in Fig. 17.
The decomposition of the antisymmetrized tensor powers can now be
written down by inspection of the dimensions involved. The module
{6} {6} is a 36-dimensional vector space, which breaks up into a sym-
metric part with (6 x 7)/(l x 2) = 21 dimensions and an antisymmetric
part with (5 x 6)/(l x 2) = 15 dimensions. From these dimensional con-
siderations and the Clebsch-Gordan series decomposition given previously,
it is clear that the decomposition of the antisymmetric part is given by
The antisymmetrized cube A 3 {6} of the fundamental module is con-

tained in
Since its dimension is (4 x 5 x 6)/(l x 2 x 3) = 20, its decomposition

can only be given by
By similar use of dimension counting and Clebsch-Gordan series, the

remaining antisymmetrized tensor powers of the fundamental module are
found to be given by
and all higher powers are zero. From this we see that the basic modules of
C3 can still be constructed from the exterior powers of the fundamental
module M, but some additional process must be used to separate off the
accompanying one- and six-dimensional modules.
3.18 SPINOR ANALYSIS AND CLIFFORD ALGEBRAS
The defining module M of an orthogonal Lie algebra can be used to

construct all of the basic modules except the spinor and semispinor modules.
Thus for the constructive representation theory of these algebras, we must
supplement the methods of tensor analysis with those of spinor analysis.
To complete the picture, we now discuss these basic spinor and semispinor
modules over the orthogonal Lie algebras. For simple Lie algebras of
type Bj, there is a single basic spinor module, while for simple Lie algebras
of type D,, there are two basic semispinor modules. The Dynkin diagrams
of the basic spinor and semispinor modules are given in Fig. 19.
Spinors were first discovered by Cartan in 1913 as a byproduct of his
classification of all simple Lie groups. Later, these spinor representations
became famous in the special case of the Lie algebra A{ w B through their
role in the Pauli theory of electron spin. To construct the spinor modules,
FIG. 19. Dynkin diagrams for the basic spinor and semispinor modules over orthogonal Lie
algebras B, and D,
we introduce the Clifford algebra of the defining module M of the orthogonal

Lie algebra L. This module M may be regarded as a module over the group
SO(n, R) or over the corresponding complex Lie algebra L. We recall that M
is an ^-dimensional complex vector space equipped with a nonsingular
symmetric bilinear form (x, y). The orthogonal Lie algebra L is the complex
Lie algebra of linear operators on M which are antisymmetric with respect
to this bilinear form. If n is an odd integer, n = 21 + 1 with / ^ 1, then L is
BI, and if n is an even integer, n = 21 with / ^ 1, then L is D,.
Clifford algebras were originally discovered as generalizations of quater-
nions [62]. The Clifford algebra related to the Lorentz group SO(3, 1;R)
is the Dirac gamma algebra, which is widely used today in relativistic
quantum mechanics [103], [188]. Since Clifford algebras are determined
abstractly by their generators and relations, we may construct them ex-
plicitly via tensor algebras [56]. Let K be the ideal in the tensor algebra
T(M) over the module M generated by all elements of the form
where z e M. The quotient algebra
is an associative algebra with unity called the first Clifford algebra of the
module M with respect to the bilinear form (x,y). Since the composition of
the canonical mapping of T(M) onto C(M) with the inclusion mapping of M
into T(M) yields a one-to-one mapping, we may identify M with its image
in C(M). The Clifford algebra C(M) obtained in this way satisfies the follow-
ing universality property. If A is an associative algebra with unity, then any
linear mapping y:M -> A which satisfies
can be extended to a unique homomorphism from C(M) into A.

To study the properties of the Clifford algebra, it is convenient to
describe the algebra in terms of generators and relations. Denoting the
product in C(M) of elements x and y in M by xy, we have the Jordan relations
for all x, y e M . The generators and relations for a Clifford algebra take the
especially simple form
if we introduce an orthonormal basis x x , , XB in M. We can then show

that the dimension of the first Clifford algebra is given by dim C(M) = 2"
since x? x'nn, where r,- = 0, 1, is a basis for the Clifford algebra.
We may note that the Jordan relations are invariant under replacement
of all vectors in M by their negatives. Consequently, the mapping z - z
in M induces an involutive automorphism of the Clifford algebra C(M).
Denoting the image of an element t e C(M) under this automorphism by t*,
we have t** = t, and z* = z for all ze M. If we split C(M) into two parts
corresponding to the eigenvalues 1 of this automorphism, then the part
C + (M) corresponding to the eigenvalue + 1 is a subalgebra of C(M) called
the second Clifford algebra [10] .
We can clarify the structures of the first and second Clifford algebras
somewhat as follows. For each vector z M , we introduce a linear operator
a(z) on C(M) by
By direct computation, one verifies that these operators anticommute, that

is, they satisfy
for all z, z' e M. If we let Mk be the subspace of C(M ) spanned by all elements
of the form a(zt) a(zk) 1, then
In particular, we note that
and hence,
Since the operators a(z;) anticommute, the expression a(z t ) a^zj 1 is

completely antisymmetric in the vectors z l 5 - - - , z k . The mapping
z, A A z k i- a ( z , ) - - - a ( z k ) - 1 establishes a vector-space isomorphism
between the exterior power A k M and the subspace M k , and hence
dimM k = (). Since
satisfies t* = ( l)kt, it should be apparent that the second Clifford algebra

is given by
In other words, the second Clifford algebra is spanned by the even elements
of the first Clifford algebra, and it follows that it is a subalgebra with
For example, the first Clifford algebra for the Lie algebra Al w B1
corresponding to the ordinary rotation group S0(3, R) is generated by three
elements xltx2 and x3 satisfying the Jordan relations. In this case, the
first Clifford algebra C(M) is an eight-dimensional algebra spanned by the
elements I,x1,x2,x3,xlx2,x2x3,x1x3 and xlx2x3. The second Clifford
algebra C + (M) is the four-dimensional subalgebra of C(M) spanned by
I , / = x1x2, j = x2x3 and k = xlx3. Since ij and k satisfy i2 = j2
k2 = 1, and ij = ji = k, jk = kj= i, and ki = ik = j, we may
identify C + (M) as the complex quaternion algebra.
Since the Clifford algebra C(M) is an associative algebra, we may also
regard it as a Lie algebra C(M)L under commutation. To see that the Lie
algebra L itself may be regarded as a subalgebra of the Lie algebra C(M)L,
we argue as follows. For all x, y and z in M we may obtain the relation
from the Jordan relation by writing the term 4(y, z)x on the right-hand side,
for instance, as
An immediate corollary of this formula is that
and hence, we also find
[M 2 ,M 2 ] = [M 2 ,[M,M]] c [[M2,M],M] c [M,M] = M2
by using the Jacobi identity. Hence, M2 is a subalgebra of the Lie algebra

C(M)L, and we shall argue that M 2 w L. For each ae M2, if we define a
linear operator f(a) on M by
then we have
for all x and y in M, and hence, f(a) e L. Since one may show that /: M 2 - L
is a one-to-one Lie algebra homomorphism, it follows that
Since the orthogonal Lie algebra L has the same dimension, dimL = ( 2 ),
the mapping / is onto, and this establishes the isomorphism of M2 with L.
The simply-connected Lie groups corresponding to the compact real
forms of BI and Dt are not the orthogonal groups S0(n, R), but their covering
groups Spin(n). To obtain these spin groups from the Clifford algebra,
we consider the group G of all invertible elements weC(M) such that
uzu~l eM for all ze M. The spin group Spin (ri) is then the identity com-
ponent of the subgroup of G consisting of all elements whose left regular
representation restricted to any ideal of C(M) has determinant one [54], [55].
To obtain the spinor and semispinor modules, we must distinguish the
case n = 21 + 1 with L = B, from the case n = 21 with L = D,. For odd n,
the first Clifford algebra C(M) is a semisimple associative algebra, isomorphic
to the direct sum of two copies of C + (M). The second Clifford algebra
C + (M) for odd n is a simple associative algebra, isomorphic to the algebra
lin (N) of linear operators on some vector space N. This vector space N
of dimension 2' will turn out to be the basic spinor module over Bl. We may
write C + (M) lin(N) N N*, and since the basic spinor module N
also happens to be self-dual, this simplifies to C + (M) N N. Thus, for
the case n odd, the second Clifford algebra C+(M) is isomorphic, as a
module over L, to the tensor square N N of the basic spinor module N.
For even n, the first Clifford algebra C(M) itself is simple, and thus is iso-
morphic to the algebra of linear operators on some vector space of dimension
2'. This vector space may be identified with the direct sum Nv + N2 of the
two basic semispinor modules N x and AT2 over >,, each having dimension
2'"1. While the semispinor modules themselves are self-dual only for even /
and each is the dual of the other for odd /, their direct sum Nv + N2 is
self-dual in either case. Thus, for even n we may write the first Clifford algebra
as
where N = N1 + N2. The second Clifford algebra C + (M) for even n is not
simple, but only semisimple, being the direct sum of two simple ideals,
namely,
We shall give a detailed explanation of these constructions only for

the case that n is odd. The basic spinor module N appears, roughly speaking,
as the square root of the second Clifford algebra C + (M) N N. In
order to extract this square root explicitly, it is convenient to introduce a
particular basis for the fundamental module M over the orthogonal Lie
algebra L = Bt . There is a basis x0 , x j , , x / for M such that
and
for all i,j = 1, , /. Writing
we have
for all /,;' = ! , - , /. The elements ut = x0Xi and vt = x 0 x_ f , where

i = 1, , / , generate C+(M) and satisfy
A basis for C + (M) consists of the set of elements
where / ! < < ir and ^ < <js. The second Clifford algebra then
factorizes as a product C+(M) = UV, where U and K are generated by the
w's and y's respectively, together with 1. The subspace N l/-( u i ty)
c= C + (M) is a 2'-dimensional left ideal of C + (M). Hence M2N <=. N, and
we may regard N as a module over M2 L.
We may illustrate this construction of the spinor module for the simplest
case, corresponding to the Lie algebra A^ B. The second Clifford algebra,
which in this case is the quaternion algebra, may be generated from the
elements
and
The algebra C + (M), spanned by !,, and MU, factorizes as 17F, where U
is spanned by 1 and u, while V is spanned by 1 and v. The two-dimensional
spinor module N = Uv is spanned by v and wy. The factorization of the
second Clifford algebra,
for the Lie algebra A t corresponds to the Clebsch-Gordan series {2} {2}
3.19 Tensor Operators 137
3.19 TENSOR OPERATORS
The theory of Lie groups and their representations enters in many

problems in physics. We briefly discuss a few simple applications to show
how Lie groups can arise and to illustrate some useful types of computational
methods. The most important of these computational tools are the Clebsch-
Gordan and Racah coefficients and the concept of a tensor operator.
The Clebsch-Gordan coefficients arise in connection with explicit
constructions of the Clebsch-Gordan series for irreducible Lie modules.
If P is an irreducible submodule included in a direct sum decomposition
of M N, then there are certain canonical homomorphisms injecting P
into M N and projecting P out of M N. The matrix elements of these
injections and projections with respect to some standard choice of module
bases are called the Clebsch-Gordan coefficients.
The theory of irreducible tensor operators constitutes an important
practical calculus for problems in atomic and nuclear physics [71], [197],
[252]. Although the concept of a tensor operator was originally introduced
only for the group SU(2), the definition is easily generalized to the case of an
arbitrary Lie algebra L. If Ml and M2 are modules over L, then we may regard
the space lin (M t , M2) of all linear mappings from Mt into M 2 as a module
over L. To do so, we define the module product of an element / 6 L and a linear
mapping relin(M 1 ,M 2 ) to be the linear mapping /telin(M 1 ,M 2 ) which
maps any xeMl into (lt)x = l(tx) - t(lx)eM2. We call any submodule
Tc= lin (Mi, M2) a tensor operator module, and its elements are called
tensor operators. There is a natural homomorphism from T ^ M ^ into
M 2 , and M2 is isomorphic to a submodule of the tensor product T M t .
The matrix elements of the tensor operators are related to the Clebsch-
Gordan coefficients which project T<g)M x onto M2 according to the
Wigner-Eckart theorem [70], [87].
The simplest example of a tensor operator module is the trivial case for
which It = 0 for all / e L and all t T. In this case
for all x e Ml so that each element t e T is a module homomorphism, or as

one sometimes says, a scalar operator. Schur's lemma can be applied to give
the structure of the module homomorphisms. In particular, if M\ and M2
are irreducible modules, then every module homomorphism is a multiple
of a given isomorphism between M^ and M 2 , if one exists, and is zero if not.
Racah coefficients typically arise in connection with recoupling prob-
lems involving tensor products of three or more Lie modules, or in compu-
tations involving tensor operators. Unlike the Clebsch-Gordan coefficients,
the Racah coefficients are invariants which do not depend on the choice
of module bases. For the calculation of the Racah coefficients, graphical
techniques prove to be useful [3].
Lie groups sometimes arise in a physical problem because of the

presence of a special dynamical symmetry. This is especially true for mathe-
matical models which owe their solvability to the presence of such a sym-
metry. One interesting example is the classical Kepler problem of describing
the motion of a planet about the sun, the planet being subject to a central
gravitational force inversely proportional to the square of its distance
from the sun. The solution of this problem leads to the familiar result that
there are two possible types of motion [202]. One possibility is a periodic
motion in a bounded orbit, the planet tracing out an elliptical path with the
center of mass at one focal point, while the other may be described as a
scattering process with a parabolic or hyperbolic trajectory. The Hamiltonian
for the system is
where p is the relative momentum, r is the distance between the sun and the
planet, m is their reduced mass, and k is a constant. Due to the rotational
invariance of the Hamiltonian, the angular momentum
is conserved. The three components Lx, Ly and Lz of the angular momentum

form a basis for the real Lie algebra of the three-dimensional rotation group.
We may recall that the Poisson bracket relations are
where sijk is the Levi-Civita tensor which is +1 (1) if (i,j, k) is an even

(odd) permutation of (1, 2, 3) and 0 otherwise. A less obviously conserved
quantity is the Runge-Lenz vector, defined by
It is a vector with dimensions of angular momentum pointing along the

major axis of the elliptic orbit of the planet. The magnitude of the Runge-
Lenz vector is equal to
times the eccentricity of the ellipse.

Some further Poisson brackets relevant to the Kepler problem may
be worked out by using elementary facts from the theory of tensor operators.
The method we use is taken from quantum mechanics, but it applies equally
well to the classical case [74]. The analogue of ihe concept of a tensor
operator in classical mechanics is a tensor dynamical variable, which we
shall illustrate only for the case of the rotation group.
3.19 Tensor Operators 139
A dynamical variable S is called a scalar dynamical variable if its Poisson

bracket with the angular momentum vanishes :
A set of three dynamical variables V = (Vx, Vy, Vz) is called a vector dynamical
variable if the Poisson bracket relations
hold. The dot product of two vector dynamical variables is a scalar dynamical
variable, while their cross product is a vector dynamical variable. Since
r and p are vector dynamical variables, it follows immediately that L x p
is also a vector dynamical variable. Hence the Runge-Lenz vector is a vector
dynamical variable, and we have
The Runge-Lenz vector further satisfies
For the bounded orbits, the system is invariant under a certain four-dimen-
sional rotation group, while the symmetry group appropriate for the
scattering trajectories turns out to be a Lorentz group. Since the energy is
negative for the bounded orbits, we have in this case simply
The vectors L and A together generate a Lie algebra which is the direct
sum of two ideals, each isomorphic to so(3, R). This algebra may be identified
with
To verify this, it is convenient to introduce the quantities
These satisfy the Poisson bracket relations
Thus one of the simple ideals is generated by the elements J + and the other
by J~.
Essentially the same Kepler problem symmetry comes up again via
the correspondence principle in the quantum mechanical description of the
hydrogen spectrum. Here the problem is to solve the Schrodinger equation
for a nonrelativistic electron moving about a proton, subject to an electro-

static force satisfying an inverse-square law [254] . Formally, the Schrodinger
equation is simply an eigenvalue problem for a linear operator called the
Hamiltonian. The eigenvalues of the Hamiltonian operator are energy
levels corresponding to the states of the system which are represented by
the corresponding eigenvectors. In the hydrogen problem, one obtains a
discrete spectrum, corresponding to the classical bounded orbits, and a
continuous spectrum, corresponding to scattering.
An eigenvalue is called degenerate when the dimension of the space
spanned by the corresponding eigenvectors is greater than one. In this case,
more than one state corresponds to the given energy level. If the Hamiltonian
does not depend explicitly on the time, the existence of conserved quantities
is directly related to the degeneracy of the spectrum. In that case, the con-
served dynamical variables correspond to operators which commute with
the Hamiltonian. The space spanned by the eigenvectors which correspond
to a given eigenvalue of the Hamiltonian is invariant under the algebra of
linear operators commuting with the Hamiltonian. To put it another way,
the eigenspaces corresponding to individual energy levels are modules
over this commutant algebra [1 23].
For the bound states of the hydrogen atom, the commutant algebra
contains a set of operators corresponding to the four-dimensional rotation
group. The description of the generators L and A in the quantum mechanical
problem is analogous to the discussion given for the classical problem. In
quantum mechanics, however, we must replace L x p in the definition of
the Runge-Lenz vector by
because these operators do not commute. The SO(4, R) symmetry can be used
to analyze the degeneracy of the spectrum of atomic hydrogen, and
historically this was the way it was first done in quantum mechanics [187].
The idea is to use the two different Casimir operators, namely L2 + A2
and A L, of the Lie algebra S0(4, R). The Hamiltonian H is related to the
first Casimir invariant by
where we have chosen units in which h = 1. From the explicit formulas for
L and A, it is easy to see that the other Casimir invariant A L is always
zero. Each eigenspace of the hydrogen atom Hamiltonian is a module over
the Lie algebra 50(4, R), but not all modules are actually realized as such
eigenspaces.
In general, the irreducible modules over a given semisimple Lie algebra
can be constructed as tensor products of irreducible modules over its
simple ideals. Explicitly, if a Lie algebra L is the direct sum of two ideals
L = L! L2 and if M1 and M2 are irreducible modules over Ll and L2,
respectively, then we may regard M x M2 as an irreducible module over L.
3.20 Charge Algebras 141
To specify the action of L on M t M 2 it is enough to say how the Lie

algebra elements act on tensor products x x x 2 of vectors x t M1 and
x2 e M 2 . If / = /i + /2 is the unique decomposition of an element / G L
as the sum of an element / t 6 Ll and an element /2 G L 2 , we set
For the hydrogen atom, the irreducible modules can be specified by the
eigenvalues of the Casimir operators (J + ) 2 and (J~) 2 . Since
only those modules occur for which these eigenvalues are the same. Since
and since
we find with n = 2j + 1,
Thus the different irreducible modules correspond to the different energy

levels of the hydrogen atom. The irreducible module corresponding to the
wth energy level has dimension (2j + I)2 = n2. The degeneracy of the nth
energy level must be some multiple of n2, and it would be n2 if the electron
had no spin. The spin of the electron contributes another factor of 2 to
the degeneracy, bringing it up to 2n2.
The symmetry of the hydrogen atom is destroyed by any small
perturbing forces, but the rotation group does not thereby lose its value.
Perturbations of the Hamiltonian, such as are produced by the presence of
an additional electric or magnetic field, may cause the eigenvalues to separate,
so that the degeneracy disappears [167], [179]. The different states then
correspond to distinct energy levels, causing what had previously been
single lines of the atomic spectrum to split into several lines. Pauli used the
S0(4, R) symmetry to obtain the splitting of the hydrogen atom energy levels
in a uniform magnetic field (Zeeman effect) and in a uniform electric field
(Stark effect) [201]. Biedenharn has also advocated using the S0(4, R)
symmetry, which he calls the geometrization of the Coulomb field, to study
the no-energy-loss Coulomb excitation of nuclear levels [31]. Biedenharn
discusses the use of the Wigner-Eckart theorem to express matrix elements
of tensor operators in terms of the Clebsch-Gordan coefficients for S0(4, R).
3.20 CHARGE ALGEBRAS

The developments of the more detailed methods of representation theory
have usually come from the demands of various specific applications. In
particular, a good deal of activity in the area of computational representation
theory has been stimulated by the needs of atomic, nuclear, and most
recently, also elementary particle physics. Since one may anticipate that in
the future there will also be many more areas in which these methods will
find application, it is likely that this field will continue to grow for some time
to come.
We discussed earlier how Lie algebras arise in the study of conservation
laws in classical and quantum mechanics. The Lie algebra of all conserved
quantities is usually infinite-dimensional since this algebra is closed not
only under the Lie product, but also under ordinary multiplication. One
can frequently obtain useful results by working with a finite-dimensional
subalgebra. For example, such finite-dimensional Lie algebras often arise
naturally as algebras of additive quantities. An additive quantity for a system
of particles is a quantity which is a sum of terms, one for each particle,
such that the term for a given particle depends only on the nature and
coordinates of that particle. Additive quantities in physics are generally
divided into two groups, those of kinematical origin, like energy and momen-
tum, and those which are not of kinematical origin. The latter are often
called generalized charges because the ordinary electric charge is an example
of just such an object. A conserved charge is one which does not depend
explicitly on the time and which commutes with the Hamiltonian. Besides
the electric charge, there is another generalized charge which is always
conserved, called the baryon number. Barring the existence of certain
hypothetical particles called quarks, the electric charge and baryon num-
ber of every known particle in physics is an integer, positive, zero or
negative.
It is widely believed that the generalized charges form a Lie algebra,
and that the set of conserved charges is a subalgebra of this Lie algebra.
Some evidence for this idea comes from the description of additive quantities
within the formalism of quantum field theory. Here we consider only the case
that all particles can be specified by means of a discrete label, which would
be the case, for instance, if they were located at fixed positions and not allowed
to move. One introduces for each type of particle a set of creation and annihi-
lation operators acting on the Hilbert space of states. The creation operators
a\, , al are required to be the Hermitian conjugates of the set of annihi-
lation operators al, , an. If \l/ is a vector in Hilbert space corresponding
to some quantum state, then the vector a]\l/ corresponds to another state
differing from the original state by the addition of a single particle of type /
Similarly, the vector afi corresponds to a state obtained by removing a
single particle of type j. These operators satisfy a certain canonical set of
commutation or anticommutation relations, depending on the spin of the
particle [43], [189].
For particles with integer spin, called bosons, they satisfy commutation
relations identical to those written down for harmonic oscillators:
3.20 Charge Algebras 143
One of the main reasons why the theory of harmonic oscillators is so funda-
mental to quantum mechanics is just because it underlies this formal
apparatus of second quantization. The creation and annihilation operators
are to a large extent determined by their commutation relations. Formally
one may regard the Hilbert space as a module over the oscillator algebra
defined by these commutation relations. If the whole Hilbert space can be
generated from a single vacuum state, for example, we obtain the Fok
representation of this oscillator algebra [88]. Von Neumann showed that
for n < oo, the Fok representation is the only irreducible representation
of the oscillator algebra [235]. For n < oo, every representation of the oscil-
lator algebra is a direct sum of Fok representations, but for n = oo the
situation becomes more complicated [51].
For half-integer spin particles, called fermions, these creation and
annihilation operators satisfy anticommutation relations,
where
The physical significance of these relations is contained in the Pauli exclusion

principle, according to which no two fermions can occupy the same quantum
state.
The Lie algebra of all generalized charges is spanned by the elements
ajaj. One can treat the case of fermions and the case of bosons simultaneously,
since in either case one can verify the commutation relations
Every generalized charge is an operator of the form Q = . â/CyO,-, where

the c^ are complex numbers. This formula, which we may rewrite as
Q = a f Ca, using an obvious matrix notation, defines a homomorphism
from the Lie algebra of n x n matrices to the charge Lie algebra. If also
R = a^Da, then from the commutation relations for ajo,- with ala{ we find
thus verifying that the map C i-+ a^Ca is a Lie algebra homomorphism.
It would be desirable to have a more direct argument that additive
quantities form a Lie algebra without appealing to the full apparatus of
quantum field theory. We sketch such an argument, which is based on the
notion of asymptotic products of state vectors. In order to say definitely
that a system is made up of a certain number of particles, it is necessary
that these particles be separated far enough apart. Otherwise their inter-
actions may be strong enough to annihilate some of the particles or to create
new particles. The state vector of a well-separated composite system may
be described as a product of the state vectors of its component clusters [113].
In this case, the statement that A is an additive quantity may be formally

interpreted as the condition
It is then easily seen that the commutator of two additive quantities is again
an additive quantity.
Lie algebras of conserved generalized charges have recently been
studied in elementary particle physics. Elementary particles engage in four
distinct classes of interactions, known as the strong, electromagnetic, weak
and gravitational interactions [59] , [253] . The strong interactions possess
greater symmetry than the others, and it is here that Lie group theory enters
particle physics most directly. Not all particles engage in strong interactions ;
those that do are called hadrons, while those that do not include the leptons
(neutrino, electron and muon) and their antiparticles, as well as the photon
and the (hypothetical) graviton. The hadrons may be further classified by
their value of the baryon number. Hadrons having baryon number 0,
1 or - 1, for example, are called mesons, baryons and antibaryons, respectively.
All known mesons have integer spin, while all known baryons have half-
integer spin.
Because the underlying dynamics of hadrons is poorly understood,
the use of symmetry considerations has been particularly valuable. Experi-
mentally, strong interaction symmetries are manifested most strikingly in
their charge and mass spectra. The earliest evidence of this sort is the near
equality of the proton and neutron masses, leading to the theory of isotopic
spin [119]. The state vectors for the proton and neutron span a two-dimen-
sional module over SU(2), called the nucleon isotopic-spin doublet. The
isotopic spin operators themselves form a representation of the corre-
sponding Lie algebra. More recently, it was found that the SU(2) symmetry
can be extended to an approximate SZ/(3) symmetry [27], [100], [217]. The
proton and neutron, together with the lambda A, sigma E + ~ and xi
hyperons E S~,span an eight-dimensional module over Sl/(3), called the
baryon octet. The strong interactions do not preserve the SU(3) symmetry,
but break it in a way which allows many of the properties of hadrons to be
predicted. For example, the masses of the baryons are not equal, but satisfy
a simple mass formula
which can be related to the symmetry-breaking mechanism [186]. Besides

the baryon octet, there are several other hadron multiplets which satisfy
mass formulas consistent with 5(7(3) symmetry. The SC/(3) symmetry is
useful not only for the strong interaction properties of the hadrons, but also
for their properties with respect to the electromagnetic and weak interactions.
While these latter interactions preserve the SU(3) symmetry even less than
the strong interactions do, the breaking mechanisms are again very definite
and many consequences can be deduced.
3.21 Clebsch-Gordan Coefficients 145
3.21 CLEBSCH-GORDAN COEFFICIENTS
The derivation of relations like the hadron mass formulas for SU(3)
involves somewhat more complicated computational techniques, especially
methods for calculating Clebsch-Gordan coefficients. To illustrate these
methods of representation theory, we shall work out one example for the
group SC/(3). The interested reader should note that the computation of
SU(2) and SU(3) Clebsch-Gordan coefficients is widely discussed in the
recent literature, and both tables and computer programs are available
[35], [58], [60].
The eight-dimensional representation whose Dynkin diagram is given
as in Fig. 20 plays an important role in applications. The Clebsch-Gordan
series
FIG. 20. Dynkin diagram of eight-dimensional SU(3) representation
can be obtained either by the use of characters or by using the Young

tableau technique. The two octets in the decomposition can be distinguished
readily because one is symmetric and the other antisymmetric :
We shall work out the Clebsch-Gordan coefficients for the symmetric

coupling of {8} {8} to {8} . For the antisymmetric coupling one can
perform a similar analysis, or one can note that the antisymmetric coupling
coefficients are related directly to the structure constants of the Lie algebra.
As a basis for the octet module, we can choose vectors ul,u2,ul2,vl,v2,
vl2,wl and vv 2 , corresponding respectively to the basis vectors el,e2,el2,
f i /2 >/i2 > î anc* h2 of the Lie algebra. The action of the Lie algebra on the
octet module can be written down immediately by inspecting the commuta-
tion relations for A2, which we wrote down earlier. For example, since
[î > e2\ e\2* we can immediately write
We list now all module products that we need :

Let us use primes to distinguish the three octet modules that are involved in
our computation of the reduction
By inspecting the weight systems, we can write down the general form for
the highest weight vector of the module (8"}:
All that we have used here is the rule
The coefficients A, B and C can be determined by making use of the fact that
MI 2 is an extreme vector: eû"^ = e2u'[2 = 0. We obtain, by a simple com-
putation,
and thus,
The overall normalization is left undetermined, since we have not specified

any particular normalization for our basis, and we might as well choose
B = 1, say, so that we obtain
Thus we obtain, for the highest weight vector,
The remaining basis vectors of the module (8"} can be obtained by applying
the lowering algebra to this highest weight vector. For example, the equation
[/2' e i2l ~e\ allows us to write down
Using our formula for u"l2, we then can immediately calculate u'[, obtaining
An interesting feature of our computation of Clebsch-Gordan coeffi-

cients in this example is that only integers appear. The reason that it is
possible for us to obtain integer Clebsch-Gordan coefficients is essentially
that we have not used the normalization conventions which are customarily
imposed [72]. Being able to work just with integers is extremely useful for
3.21 Clebsch-Gordan Coefficients 147
programming automatic electronic digital computers to compute Clebsch-

Gordan coefficients. A general proof that it is always possible to choose
module bases for which the Clebsch-Gordan coefficients are integers for
any simple Lie algebra has been given recently by Chevalley [57], [212].
It should be noted that in the course of solving the equations which
determine the extreme subspace in a given reduction, it sometimes happens
that we obtain more than one extreme vector with the same weight. This is
sometimes referred to as the multiplicity problem for Clebsch-Gordan
coefficients. In the preceding discussion, for example, there are two octet
submodules in the tensor product {8} {8}, which we could distinguish
on the basis of symmetry. But even if there were no symmetry to distinguish
such isomorphic submodules, we could still find a basis for the extreme sub-
space. Finding such.a basis using integer mode arithmetic amounts to
solving a set of Diophantine linear equations, as our example shows. The
solution of such a set of equations is equivalent to finding a basis for a
finitely generated Abelian group. This can be done automatically on a
computer by the use of an algorithm published recently by D. A. Smith
[213], [220]. Therefore, at least in principle, all multiplicity problems can
be resolved in the calculation of Clebsch-Gordan coefficients for any
simple Lie algebra.
In practice, however, this method is seldom used by physicists because
they are usually interested in unitary representations. For the unitary
representations, it is often more useful to employ an orthonormal
basis, even though this means giving up integer Clebsch-Gordan coefficients.
Although there is no uniform method for finding such orthonormal bases
which applies to any simple Lie algebra, much work has been done for
various special cases. In the case of SU(n), for example, much work has been
done to calculate Clebsch-Gordan coefficients using a basis discovered by
Gelfand and Zetlin [153].
BIBLIOGRAPHY
[1] R. ABRAHAM, Foundations of Mechanics. A Mathematical Exposition of Classical

Mechanics with an Introduction to the Qualitative Theory of Dynamical Systems and
Applications to the Three-Body Problem, Mathematical Physics Monograph Series,
W. A. Benjamin, New York, 1967.
[2] I. D. ADO, The representation of Lie algebras by matrices, Uspekhi Mat. Nauk (22) 3
(1947), no. 6, pp. 159-173; English transl., Amer. Math. Soc. Transl. no. 2,1949.
[3] V. K. AGRAWALA AND J. G. F. BELINFANTE, Graphical formulation ofrecoupling theory
for any compact group, Ann. Physics, 49 (1968), pp. 130-170.
[4] , Weight diagrams for Lie group representations: A computer implementation of
Freudenthal's algorithm in ALGOL and FORTRAN, BIT, 9 (1969), pp. 301-314.
[5] , An algorithm for computing SU(ri) invariants, BIT, 11 (1971), pp. 1-15.
[6] C. M. ANDERSEN, Clebsch-Gordan series for symmetrized tensor products, J. Mathematical
Phys., 8 (1967), pp. 988-997.
[7] J. P. ANTOINE, Irreducible representations of the group SU(3), Ann. Soc. Sci. Bruxelles
Ser. I., 77 (1963), pp. 150-162.
[8] J. P. ANTOINE AND D. R. SPEISER, Characters of the irreducible representations of the simple
groups. I, J. Mathematical Phys., 5 (1964), pp. 1226-1234.
[9] 5 Characters of the irreducible representations of the simple groups. II: Application
to classical groups, Ibid., 5 (1964), pp. 1560-1572.
[10] E. ARTIN, Geometric Algebra, Interscience Tracts in Pure and Applied Mathematics,
vol. 3, Interscience, New York, 1957.
[11] E. W. ASLAKSEN AND J. R. K.LAUDER, Unitary representations of the affine group, J. Mathe-
mathetical Phys., 9 (1968), pp. 206-211.
[12] , Continuous representation theory using the affine group, Ibid., 10 (1969), pp.
2267-2275.
[13] H. BACRY, Position and polarization operators in relativistic and nonrelativistic mechanics,
Ibid., 5 (1964), pp. 109-111.
[14] G. E. BAIRD AND L. C. BIEDENHARN, JR., On the representation of the semisimple Lie
groups. II, Ibid., 4 (1963), pp. 1449-1466.
[15] G. A. BAKER, Degeneracy of the n-dimensional isotropic harmonic oscillator, Phys. Rev.,
103 (1956), pp. 1119-1120.
[16] H. F. BAKER, Alternants and continuous groups, Proc. London Math. Soc. (2), 3 (1904),
pp. 24-47.
[17] V. BARGMANN, Irreducible unitary representations of the Lorentz group, Ann. of Math.
(2), 48 (1947), pp. 568-642.
[18] , On unitary ray representations of continuous groups, Ibid., (2), 59 (1954), pp. 1^46.
[19] , Note on Wigner's theorem on symmetry operations, J. Mathematical Phys., 5
(1964), pp. 862-868.
[20] V. BARGMANN AND M. MOSHINSKY, Group theory of harmonic oscillators. Part I: The
collective modes, Nuclear Phys., 18 (1960), pp. 697-712.
[21] D. W. BARNES, Nilpotency of Lie algebras, Math. Z., 79 (1962), pp. 237-238.
149
150 Bibliography
[22] D. W. BARNES, Lattice isomorphisms of Lie algebras, J. Australian Math. Soc., 4 (1964),
pp. 470-475.
[23] , On Carton subalgebras of Lie algebras, Math. 2., 101 (1967), pp. 350-355;
Erratum, Math. Z., I l l (1969), p. 168.
[24] R. E. BECK AND B. KOLMAN, Generation of the Weyl group on a computer, J. Computa-
tional Phys., 7 (1971), pp. 346-353.
[25] , Computer approaches to the representations of Lie algebras, J. Assoc. Comput.
Mach., to appear.
[26] , A computer implementation of Freudenthaf s multiplicity formula, to be published.
[27] R. E. BEHRENDS, J. DREITLEIN, C. FRONSDAL AND B. W. LEE, Simple groups and strong
interaction symmetries, Rev. Modern Phys., 34 (1962), pp. 1-40.
[28] J. G. F. BELINFANTE, B. KOLMAN AND H. A. SMITH, An introduction to Lie groups and
Lie algebras, with applications, SIAM Rev., 8 (1966), pp. 11-46.
[29] , An introduction to Lie groups and Lie algebras, with applications. II: The basic
methods and results of representation theory, Ibid., 10 (1968), pp. 160-195.
[30] J. G. F. BELINFANTE AND B. KOLMAN, An introduction to Lie groups and Lie algebras,
with applications. Ill: Computational methods and applications of representation
theory, Ibid., 11 (1969), pp. 510-543.
[31] L. C. BIEDENHARN, JR., Wigner coefficients for the /?4 group and some applications,
J. Mathematical Phys., 2 (1961), pp. 433-441.
[32] L. C. BIEDENHARN, JR. AND W. J. HOLMAN, III, Complex angular momenta and the groups
SU(1, /) and SU(2), Ann. Physics, 39 (1966), pp. I^t2.
[33] G. BIRKHOFF, Lie groups isomorphic with no linear group, Bull. Amer. Math. Soc., 42
(1936), pp.882-888.
[34] , Hydrodynamics: A Study in Logic, Fact, and Similitude, Princeton Univ. Press,
Princeton, N.J., 1950.
[35] R. BIVINS, N. METROPOLIS, M. ROTENBERG AND J. K. WOOTEN, JR., The 3 j and 6 j
Symbols, The M.I.T. Press, Cambridge, Mass., 1959.
[36] A. BOREL, Topology of Lie groups and characteristic classes, Bull. Amer. Math. Soc.,
61 (1955), pp. 397-432.
[37] R. BRANTON, A generalization of the fundamental theorem ofprojective geometry, Doctoral
thesis, University of Pennsylvania, Philadelphia, 1970.
[38] F. BRICKELL AND R. S. CLARK, Differentiable Manifolds: An Introduction, Van Nostrand-
Reinhold, London, 1970.
[39] H. C. BRINKMAN, Applications of Spinor Invariants in Atomic Physics, North Holland,
Amsterdam, 1956.
[40] R. BROCKETT, System theory on group manifolds and coset spaces, SIAM J. Control,
10 (1972), pp. 265-284.
[41] G. BROWN, A remark on semisimple Lie algebras, Proc. Amer. Math. Soc., 15 (1964),
p. 518.
[42] F. BRUHAT, Lectures on Lie groups and representations of locally compact groups, Lecture
notes, Tata Institute of Fundamental Research, Bombay, India, 1958.
[43] N. BURGOYNE, On the connection of spin with statistics, Nuovo Cimento, 8 (1958), pp. 607-
609.
[44] J. E. CAMPBELL, On a law of combination of operators, Proc. London Math. Soc., 29
(1898), pp. 14-32.
[45] E. CARTAN, Projective groups which do not leave any flat manifold invariant, Bull. Soc.
Math. France, 41 (1913), pp. 53-96 (in French).
[46] , The finite and continuous real simple groups, Ann. Sci. Ecole Norm. Sup., 31
(1914), pp. 263-355 (in French).
[47] P. CARTIER, Finite groups generated by reflections, Seminaire Chevalley, vol. 2, expose
14, Ecole Normale Superieure, Paris, 1957/1958 (in French).
[48] , On H. WeyTs character formula, Bull. Amer. Math. Soc., 67 (1961), pp. 228-230.
Bibliography 151
[49] H. CASIMIR, On the construction of a differential equation corresponding to the irreducible

representations of semisimple continuous groups, Proc. Konikl. Akad. Amsterdam,
34 (1931), pp. 844-846 (in German).
[50] H. CASIMIR AND B. L. VAN DER WAERDEN, Algebraic proof of the complete reducibility
of the representations of semisimple groups, Math. Ann., I l l (1935), pp. 1-12 (in
German).
[51] J. M. CHAIKEN, Number operators for representations of the canonical commutation
relations, Comm. Math. Phys., 8 (1968), pp. 164-184.
[52] K. T. CHEN, Formal differential equations, Ann. of Math. (2), 73 (1961), pp. 110-133.
[53] , Decomposition of differential equations, Math. Ann., 146 (1962), pp. 263-278.
[54] C. C. CHEVALLEY, Theory of Lie Groups, vol. I, Princeton Univ. Press, Princeton, N.J.,
1946.
[55] , The Algebraic Theory of Spinors, Columbia Univ. Press, New York, 1954.
[56] , The Construction and Study of Certain Important Algebras, Mathematical Society
of Japan, Tokyo, 1955.
[57] , On Certain Simple Groups, Tohoku Math. J. (2), 7 (1955), pp. 14-66 (in French).
[58] C.-K. CHEW AND H. C. VON BAEYER, Explicit computation of the SU(3) Clebsch-Gordan
coefficients, Nuovo Cimento, 56 A (1968), pp. 53-63.
[59] G. F. CHEW, M. GELL-MANN AND A. H. ROSENFELD, Strongly interacting particles,
Scientific American, 210 (1964), no. 2, pp. 74-93.
[60] F. CHILTON AND P. MCNAMEE, Tables of Clebsch-Gordan coefficients of SU(3), Rev.
Modern Phys., 36 (1964), pp. 1005-1024.
[61] Y. CHOW, Gel'fand-Kirillov conjecture on the Lie field of an algebraic Lie algebra, J.
Mathematical Phys. 10 (1969), pp. 975-992.
[62] W. K. CLIFFORD, Applications of Grossman's extensive algebra, Amer. J. Math., 1 (1878),
pp. 350-358.
[63] A. COHEN, An Introduction to the Lie Theory of One-Parameter Groups, Stechert, New
York, 1931.
[64] P. M. COHN, Lie Groups, Cambridge Tracts in Mathematics and Mathematical Physics,
No. 46, Cambridge Univ. Press, Cambridge, 1957.
[65] H. C. CORBEN AND P. STEHLE, Classical Mechanics, 2nd ed., John Wiley, New York,
1960.
[66] E. M. CORSON, Introduction to Tensors, Spinors and Relativistic Wave Equations, Blackie
& Son, London, 1953.
[67] H. S. M. COXETER, Discrete groups generated by reflections, Ann. of Math. (2), 35 (1934),
pp. 588-621.
[68] , The product of the generators of 'a finite group generated by reflections, Duke Math.
J., 18 (1951), pp. 765-782.
[69] H. S. M. COXETER AND W. O. J. MOSER, Generators and Relations for Discrete Groups,
Springer-Verlag, Berlin, 1957.
[70] J. R. DEROME AND W. T. SHARP, Racah algebra for an arbitrary group, J. Mathematical
Phys., 6 (1965), pp. 1584-1590.
[71] A. DE SHALIT AND I. TALMI, Nuclear Shell Theory, Academic Press, New York, 1963.
[72] J. J. DE SWART, The octet model and its Clebsch-Gordan coefficients, Rev. Modern Phys.,
35(1963), pp. 916-939; Erratum: Rev. Modern Phys., 37 (1965), p. 326.
[73] H. DE VRIES AND H. FREUDENTHAL, Linear Lie Groups, Academic Press, New York,
1969.
[74] R. H. DlCKE AND J. P. WITTKE, Introduction to Quantum Mechanics, Addison-Wesley,
Reading, Mass., 1960.
[75] P. A. M. DIRAC, Relativity quantum mechanics with application to Compton scattering,
Proc. Roy. Soc. London, A 111 (1926), pp. 405-423.
[76] , On the analogy between classical and quantum mechanics, Rev. Modern Phys.,
17 (1945), pp. 195-199.
152 Bibliography
[77] P. A. M. DIRAC, Generalized Hamiltonian dynamics, Canad. J. Math., 2 (1950), pp.

129-148.
[78] , Generalized Hamiltonian dynamics, Proc. Roy. Soc. London, A 246 (1958),
pp. 326-332.
[79] , Principles of Quantum Mechanics, 4th ed., Clarendon Press, Oxford, 1958.
[80] N. DUNFORD AND J. T. SCHWARTZ, Linear Operators, Part II. Spectral Theory: Self-
Adjoint Operators in Hilbert Space, Interscience, New York, 1963.
[81] E. B. DYNKIN, The structure of semisimple algebras, Uspekhi Mat. Nauk, 2 (1947),
pp. 59-127; English transl., Amer. Math. Soc. Transl. No. 17, 1950, reprinted in
Amer. Math. Soc. Translations Series I, vol. 9, 1962, pp. 328-469.
[82] , Semisimple subalgebras of semisimple Lie algebras. Mat. Sb., 30 (1952), pp.
349-462; English transl., Amer. Math. Soc. Translations Series II, vol. 6, 1957,
pp.111-244.
[83] , The maximal subgroups of the classical groups, Trudy Moskov. Mat. Obshch.,
1 (1952), pp. 39-166, includes supplement, Survey of the basic concepts and facts in
the theory of linear representations of semisimple Lie algebras; English transl., Amer.
Math. Soc. Transl. (2), 6 (1957), pp. 245-378.
[84] A. R. EDMONDS, Angular Momentum in Quantum Mechanics, Princeton Univ. Press,
Princeton, N.J., 1957.
[85] D. L. ELLIOTT, A consequence of controllability, J. Differential Equations, 10 (1971),
pp. 364-370.
[86] D. L. ELLIOTT AND D. J. TARN, Controllability and observability for bilinear systems,
preprint, (1971).
[87] U. FANO AND G. RACAH, Irreducible Tensorial Sets, Academic Press, New York, 1959.
[88] V. FOCK, Configuration space and Dirac's method of'quantization, Z. Physik, 75(1932),pp.
622-647.
[89] H. FREUDENTHAL, On the calculation of the characters of semisimple Lie groups. I, Nederl.
Akad. Wetensch. Proc. Ser. A, A 57(1954), pp. 369-376; Nederl. Akad. Wetensch.
Indag. Math., 16 (1954), pp. 369-376 (both in German).
[90] , On the calculation of the characters of semisimple Lie groups. II. Nederl. Akad.
Wetensch. Proc. Ser. A, A 57(1954), pp. 487-491; Nederl. Akad. Wetensch. Indag.
Math., 16 (1954), pp. 487-491 (both in German).
[91] , On the calculation of the characters of semisimple Lie groups. Ill, Nederl. Akad.
Wetensch. Proc. Ser. A, A 59(1956), pp. 511-514; Nederl. Akad. Wetensch. Indag.
Math., 18 (1956), pp. 511-514 (both in German).
[92] , Lie Groups, Yale University Lecture Notes, New Haven, Connecticut, 1961,
unpublished.
[93] F. GANTMACHER, Canonical representations of automorphisms of a complex semisimple
Lie Group, Mat. Sb., 5 (1939), pp. 101-146 (in English).
[94] , On the classification of real simple Lie groups, Ibid., 5 (1939), pp. 217-249.
[95] L. CARDING, Relativistic wave equations for zero rest-mass, Proc. Cambridge Philos. Soc.,
41 (1945), pp. 49-56.
[96] , Note on continuous representations of Lie groups, Proc. Nat. Acad. Sci. U.S.A.,
33 (1947), pp. 331-332.
[97] L. CARDING AND A. S. WIGHTMAN, Representations of the commutation relations, Ibid.,
40 (1954), pp. 622-626.
[98] I. M. GEL'FAND AND A. A. KIRILLOV, Fields associated with enveloping algebras of Lie
algebras, Dokl. Akad. Nauk SSSR, 167 (1966), pp. 503-505; English transl., Soviet
Math. Dokl.,.7 (1966), pp. 407-409.
[99] , The structure of the quotient field of the enveloping algebra of a semisimple Lie
algebra, Dokl. Akad. Nauk SSSR, 180 (1968), pp. 775-777; English transl., Soviet
Math. Dokl., 9 (1968), pp. 669-671.
[100] M. GELL-MANN AND Y. NE'EMAN, The Eightfold Way, W. A. Benjamin, New York, 1964.
Bibliography 153
[101] R. C. GLAESER AND B. KOLMAN, Lattice isomorphisms of solvable Lie algebras, J. Australian
Math. Soc., 10 (1969), pp. 266-268.
[102] R. GODEMENT, A theory of spherical functions. I, Trans. Amer. Math. Soc., 73 (1952),
pp. 496-555.
[103] R. H. GOOD, Properties of the Dirac matrices. Rev. Modern Phys., 27 (1955),
pp. 187-211.
[104] M. GOTO, On algebraic Lie algebras, J. Math. Soc. Japan, 1 (1948), pp. 29-^45.
[105] , Lattices of subalgebras of real Lie algebras, J. Algebra, 11 (1969), pp. 6-24.
[106] M. GOURDIN, Unitary Symmetries and Their Application to High-Energy Physics, North
Holland, Amsterdam, 1967.
[107] S. GREENSPAN AND R. D. RICHTMYER, Expansion of the Campbell-Baker-Hausdorff
formula by computer, Comm. Pure Appl. Math., 18 (1965), pp. 107-108.
[108] H. J. GROENEWOLD, On the principles of elementary quantum mechanics, Physica, 12
(1946), pp. 405-474.
[109] B. GRUBER, Lecture notes on multiplicities in the classical groups, Institute di Fisica
Teorica, Universita di Napoli, Naples, Italy, 1968, unpublished.
[110] , Recurrence relations for the multiplicities in the classical groups, J. Mathematical
Phys., 11 (1970), pp. 1783-1790.
[Ill] B. GRUBER AND H. J. WEBER, On the construction of weight diagrams for SO(5), Sp(4)
andG2, Proc. Roy. Irish Acad. Sect. A, 66 A (1968), pp. 31^40.
[112] K. W. GRUENBERG AND A. J. WEIR, Linear Geometry, Van Nostrand, Princeton, N.J.,
1967.
[113] R. HAAG, The framework of quantum field theory, Nuovo Cimento Suppl., 14 (1959),
pp. 131-152.
[114] P. R. HALMOS, Finite-Dimensional Vector Spaces, 2nd ed., Van Nostrand, Princeton,
N.J., 1958.
[115] M. HAMERMESH, Group Theory and its Applications to Physical Problems, Addison-
Wesley, Reading, Mass., 1962.
[116] HARISH-CHANDRA, On representations of Lie algebras, Ann. of Math. (2), 50 (1949),
pp. 900-915.
[117] , On the radical of a Lie algebra, Proc. Amer. Math. Soc., 1 (1950), pp. 14-17.
[118] F. HAUSDORFF, The symbolic exponential formula in group theory, Ber. Verh. Sachs.
Ges. Wiss. Leipzig. Math.-Phys. K.I., 58 (1906), pp. 19-^8 (in German).
[119] W. HEISENBERG, On the structure of the atomic nucleus. I, Z. Physik, 77 (1932), pp. 1-11
(in German).
[120] S. HELGASON, Differential Geometry and Symmetric Spaces, Academic Press, New York,
1962.
[121] K. HELMERS, Symplectic invariants and Flowers" classification of shell model states,
Nuclear Phys., 23 (1961), pp. 594-611.
[122] J. C. HERZ. Contribution to the algebraic theory of partial differential equations, Ann. Sci.
ficole Norm. Sup., 71 (1954), pp. 321-362.
[123] E. L. HILL AND J. M. JAUCH, On the problem of degeneracy in quantum mechanics, Phys.
Rev., 57 (1940), pp. 641-645.
[124] G. HOCHSCHILD, The Structure of Lie Groups, Holden-Day, San Francisco, 1965.
[125] W. C. HOFFMAN, The neuron as a Lie group germ and a Lie product, Quart. Appl. Math.,
25(1968), pp.423-440.
[126] H. HOPF, Vector fields in n-dimensional manifolds, Math. Ann., 96 (1926), pp. 225-250
(in German).
[127] H. HOPF AND W. RINOW, On the concept of a complete differential-geometric surface,
Comment. Math. Helv., 3 (1931), pp. 209-225 (in German).
[128] T. E. HULL AND L. INFIELD, The factorization method, Rev. Modern Phys., 23 (1951),
pp. 21-68.
[129] T. HUSAIN, Introduction to Topological Groups, W. B. Saunders, Philadelphia, 1966.
154 Bibliography
[130] E. L. INCH, Ordinary Differential Equations, Longmans, Green & Co., London, 1926;
reprinted by Dover Publications, New York, 1956.
[131] C. ITZYKSON AND M. NAUENBERG, Unitary groups: Representations and decompositions,
Rev. Modern Phys., 38 (1966), pp. 95-120.
[132] M. JACOB AND G. C. WICK, On the general theory of collisions for particles with spin,
Ann. Physics, 7 (1959), pp. 404-428.
[133] N. JACOBSON, Rational methods in the theory of Lie algebras, Ann. of Math. (2), 36
(1935), pp. 875-881.
[134] , Lectures in Abstract Algebra. Volume II: Linear Algebra, Van Nostrand, Prince-
ton, N.J., 1953.
[135] , Lie Algebras, Interscience Tracts in Pure and Applied Mathematics, No. 10,
Interscience, New York, 1962.
[136] , Lectures in Abstract Algebra. Volume HI: Theory of Fields and Galois Theory,
Van Nostrand, Princeton, N.J., 1964.
[137] A. JOSEPH, Derivations of Lie brackets and canonical quantization, Comm. Math. Phys.,
17 (1970), pp. 210-232.
[138] R. JOST, Poisson brackets (an unpedagogical lecture), Rev. Modern Phys., 36 (1964),
pp.572-579.
[139] R. E. KALMAN, Mathematical description of linear dynamical systems, SI AM J. Control,
1 (1963), pp. 152-192.
[140] A. U. KLIMYK, Multiplicities of weights of representations and multiplicities of representa-
tions ofsemisimple Lie algebras, Soviet Math. Dokl., 8 (1967), pp. 1531-1534.
[141] B. KOLMAN, Semi-modular Lie algebras, J. Sci. Hiroshima University Ser. A-I Math.,
29 (1965), pp. 149-163.
[142] , Relatively complemented Lie algebras, Ibid., 31 (1967), pp. 1-11.
[143] M. KONUMA, K. SHIMA AND M. WADA, Simple Lie algebras of rank 3 and symmetries of
elementary particles in the strong interactions, Progr. Theoret. Phys. (Japan) Suppl.,
28(1963), pp. 1-128.
[144] P. Koosis, An irreducible unitary representation of a compact group is finite-dimensional,
Proc. Amer. Math. Soc., 8 (1957), pp. 712-715.
[145] B. KOSTANT, A formula for the multiplicity of a weight, Proc. Nat. Acad. Sci. U.S.A.,
44 (1958), pp. 588-589.
[146] , A formula for the multiplicity of a weight, Trans. Amer. Math. Soc., 93 (1959),
pp. 53-73.
[147] , Lie group representations on polynomial rings, Bull. Amer. Math. Soc., 69(1963),
pp. 518-526.
[148] M. I. KRUSEMEYER, Determining multiplicities of dominant weights in irreducible Lie
algebra representations using a computer, BIT, 11 (1971), pp. 310-316.
[149] J. KUCERA, Solution in large of control problem x = (A(\ - v) + Bv)x, Czechoslovak.
Math. J., 16 (1966), no. 91, pp. 600-623.
[150] , Solution in large of control problem x - (Au + Bv)x, Ibid., 17 (1967), no. 92,
pp.91-96.
[151] E. E. LEVI, On the structure of finite and continuous groups, Atti Accad. Torino, 40 (1905),
pp. 3-17 (in Italian).
[152] L. H. LOOMIS, An Introduction to Abstract Harmonic Analysis, Van Nostrand, New York,
1953.
[153] J. D. LOUCK, Recent progress toward a theory of tensor operators in the unitary groups,
Amer. J. Phys., 38 (1970), pp. 3-42.
[154] G. LOUPIAS, M. SIRUGUE AND J. C. TROTIN, On simple Lie groups of rank 3, Nuovo Ci-
mento. 38 (1965), pp. 1303-1325.
[155] A. J. MACFARLANE, L. O'RAIFEARTAIGH AND P. S. RAO, Relationship of the internal and
external multiplicity structure of compact simple Lie groups, J. Mathematical Phys.,
8 (1967), pp. 536-546.
Bibliography 155
[156] G. W. MACKEY, Infinite-dimensional group representations, Bull. Amer. Math. Soc.,

69 (1963), pp. 628-686.
[157] S. MACLANE, Natural associativity and commutativity, Rice Univ. Studies, 49 (1963),
pp.28-46.
[158] P. A. MACMAHON, Combinatory Analysis, vol. II, Cambridge University Press, Cam-
bridge, 1916.
[159] W. MAGNUS, On the exponential solution of differential equations for a linear operator,
Comm. Pure Appl. Math., 7 (1954), pp. 649-673.
[160] A. I. MAL'CEV, On the representation of an algebra as a direct sum of its radical and a semi-
simple subalgebra, Dokl. Akad. Nauk SSSR, 36 (1942), pp. 42-45.
[161] , On semisimple subgroups of Lie groups, Amer. Math. Soc. Transl. (I), 9 (1962),
pp. 172-213.
[162] K. MAURIN, General Eigenfunction Expansions and Unitary Representations of Topo-
logical Groups, Polish Scientific Publishers, Warsaw, Poland, 1968.
[163] W. MAYER AND T. Y. THOMAS, Foundations of the theory of Lie groups, Ann. of Math.
(2), 36 (1935), pp. 770-822.
[164] G. McCARTY, Topology: An Introduction with Application to Topological Groups,
McGraw-Hill, New York, 1967.
[165] J. McCONNELL, Multiplicities in weight diagrams, Proc. Roy. Irish Acad. Sect. A, A 65
(1966), pp. 1-12.
[166] , Properties of the exceptional G2 Lie group. Ibid., A 66 (1968), pp.
72-92.
[167] A. MESSIAH, Quantum Mechanics, vol. II, North Holland, Amsterdam, 1962.
[168] W. MILLER, JR., On Lie algebras and some special functions of mathematical physics,
Mem. Amer. Math. Soc., no. 50, 1964.
[169] , Lie Theory and Special Functions, Math. Sci. Engrg. Series, vol. 43, Academic
Press, New York, 1968.
[170] D. MONTGOMERY AND L. ZIPPIN, Topological Transformation Groups, Interscience,
New York, 1955.
[171] C. B. MORREY, The analytic embedding of abstract real-analytic manifolds, Ann. of
Math. (2), 68 (1958), pp. 159-201.
[172] H. E. MOSES, Reduction of reducible representations of the Poincare group to standard
helicity representations, J. Mathematical Phys., 9 (1968), pp. 2039-2049.
[173] M. MOSHINSKY, The harmonic oscillator and supermultiplet theory. /: The single shell
picture, Nuclear Phys., 31 (1962), pp. 384-405.
[174] G. M. MUBARAKZJANOV, On solvable Lie algebras, Izv. Vyssh. Uchebn. Zaved. Mate-
matika, 1 (1963), no. 32, pp. 114-123 (in Russian).
[175] , Classification of real structures of Lie algebras of fifth order, Ibid., 3 (1963),
no. 34, pp. 99-106 (in Russian).
[176] , Classification of solvable Lie algebras of sixth order with a non-nilpotent basis
element. Ibid., 4(1963), no. 35, pp. 104-116 (in Russian).
[177] , Certain theorems on solvable Lie algebras, Ibid., 6 (1966), no. 55, pp. 95-98 (in
Russian).
[178] S. MURAKAMI, On the classification of real simple Lie algebras, Osaka J. Math., 2 (1965),
pp. 291-307 (in French).
[179] F. J. MURRAY, Perturbation theory and Lie algebras, J. Mathematical Phys., 3 (1962),
pp. 451-468.
[180] L. NACHBIN, On the finite dimensionality of every irreducible representation of a compact
group, Proc. Amer. Math. Soc., 12 (1961), pp. 11-12.
[181] M. A. NAIMARK, Normed Rings, P. Noordhoff, Groningen, the Netherlands,
1964.
[182] , Linear Representations of the Lorentz Group, Fizmatgiz, Moscow, 1958 (in
Russian); English transl., Macmillan, New York, 1964.
156 Bibliography
[183] Y. NE'EMAN, The symmetry approach to particle physics, Proc. International Conference
on Nuclear Structure, 1963, pp. 172-187; Reprinted in Gell-Mann and Ne'eman,
The Eightfold Way, W. A. Benjamin, New York, 1964, pp. 302-317.
[184} E. NORMAN AND J. WEI, Lie algebraic solution of linear differential equations, J. Mathe-
matical Phys., 4 (1963), pp. 575-581.
[185] , On global representations of the solutions of linear differential equations as a
product of exponentials, Proc. Amer. Math. Soc., 15 (1964), pp. 327-334.
[186] S. OKUBO, Note on unitary symmetry in strong interactions, Prog. Theoret. Phys., 27
(1962), pp. 949-966.
[187] W. PAULI, JR., On the hydrogen spectrumfrom the standpoint of the new quantum mechanics,
Z. Physik, 36 (1926), pp. 336-363.
[188] , Mathematical contributions to the theory ofDirac matrices, Ann. Inst. H. Poincare,
6 (1936), pp. 109-136 (in French).
[189] , The connection between spin and statistics, Phys. Rev., 58 (1940), pp. 716-722.
[190] , Continuous groups in quantum mechanics, Ergeb. d. Exakt. Naturw., vol. 37,
G. Hohler, ed., Springer-Verlag, Berlin, 1965, pp. 85-104 (Posthumous publication
of 1958 CERN Report).
[191] S. PERLIS, Theory of Matrices, Addison-Wesley, Cambridge, Mass., 1952.
[192] L. S. PONTRVAGIN, Topological Groups, 2nd ed., Gordon and Breach, New York, 1966.
[193] L. PUKANSZKY, On the Kronecker products of irreducible representations of the 2x2
real unimodular group. I, Trans. Amer. Math. Soc., 100 (1961), pp. 116-152.
[194] G. RACAH, Theory of complex spectra. IV, Phys. Rev., 76 (1949), pp. 1352-1365.
[195] , On the decomposition of tensors by contraction, Rev. Modern Phys., 21 (1949),
pp. 494-496.
[196] , Lectures on Lie groups, Group Theoretical Concepts and Methods in Elementary
Particle Physics, F. Gursey, ed., Gordon and Breach, New York, 1964, pp. 1-36.
[197] , Group theory and spectroscopy, Ergeb. d. Exakt. Naturw., vol. 37, G. Hohler,
ed., Springer-Verlag, Tracts in Modern Physics, Berlin, 1965, pp. 28-84 (Reprint
of 1951 lecture notes).
[198] D. RADHAKRISHNAN, On the Clebsch-Gordan series of a semisimple Lie algebra, J. Mathe-
matical Phys., 9(1968), pp. 2061-2063.
[199] D. RADHAKRISHNAN AND T. S. SANTHANAM, Internal multiplicity structure and Clebsch-
Gordan series for the exceptional group G(2), Ibid., 8 (1967), pp. 2206-2209.
[200] S. RAMANUJAM, Topology of classical groups, Osaka J. Math., 6 (1969), pp. 243-249.
[201] P. J. REDMOND, Generalization of the Runge-Lenz vector in the presence of an electric
field, Phys. Rev., 133 (1964), pp. 1352-1353.
[202] D. E. RICHMOND, Inverse square orbits: A simple treatment, Amer. Math. Monthly,
59 (1952), pp. 694-696.
[203] M. E. ROSE, Elementary Theory of Angular Momentum, John Wiley, New York, 1957.
[204] D. E. RUTHERFORD, Substitutional Analysis, Edinburgh University Press," Edinburgh,
1948.
[205] H. SAMELSON, Topology of Lie groups. Bull. Amer. Math. Soc., 58 (1952), pp. 2-37.
[206] T. SATAKE, On representations and compactifications of symmetric spaces, Ann. of Math.
(2), 71 (1960), pp. 77-110.
[207] R. D. SCHAFER, An Introduction to Nonassociative Algebras, Academic Press, New York,
1966.
[208] J. A. SCHOUTEN, Tensor Analysis for Physicists, Clarendon Press, Oxford, 1954.
[209] J. P. SERRE, Semisimple Complex Lie Algebras, W. A. Benjamin, New York, 1966 (in
French).
[210] J. R. SHEWELL, On the formation of quantum-mechanical operators, Amer. J. Phys., 27
(1959), pp. 16-21.
[211] D. J. SIMMS, Lie Groups and Quantum Mechanics, Lecture Notes in Mathematics, No. 52,
Springer-Verlag, Berlin, 1968.
Bibliography 157
[212] D. A. SMITH, Chevalley bases for Lie modules, Trans. Amer. Math. Soc., 115 (1965),
pp. 283-299.
[213] , A basis algorithm for finitely-generated Abelian groups, Math. Algorithms, 1
(1966), pp. 14-35.
[214] J. M. SOURIAU, Geometric Quantization, Comm. Math. Phys., 1 (1966), pp. 374-398
(in French).
[215] B. SPAIN, Tensor Calculus, University Math. Texts, No. 17, Oliver and Boyd, London, 1956.
[216] D. R. SPEISER, Fundamental representations of Lie groups, Helv. Phys. Acta, 38 (1965),
pp. 73-97.
[217] D. R. SPEISER AND J. TARSKI, Possible schemes for global symmetry, J. Mathematical
Phys., 4 (1963), pp. 588-612.
[218] T. A. SPRINGER, WeyVs character formula for algebraic groups, Inventiones Math., 5
(1968), pp. 85-105.
[219] R. STEINBERG, A general Clebsch-Gordan theorem, Bull. Amer. Math. Soc., 67 (1961),
pp. 401-407.
[220] E. L. STIEFEL, An Introduction to Numerical Mathematics, Academic Press, New York,
1963.
[221] N. STRAUMANN, On the Clebsch-Gordan series of semisimple Lie algebras, Helv. Phys.
Acta, 38 (1965), pp. 56-64.
[222] , Branching rules and Clebsch-Gordan series of semisimple Lie algebras, Ibid.,
38 (1965), pp. 481-498.
[223] R. F. STREATER, Canonical quantization, Comm. Math. Phys., 2 (1966), pp. 354-374.
[224] M. E. SWEEDLER, Hopf Algebras, W. A. Benjamin, New York, 1969.
[225] J. D. TALMAN, Special Functions: A Group Theoretic Approach, W. A. Benjamin, New
York, 1968.
[226] J. TARSKI, Partition function for certain simple Lie algebras, J. Mathematical Phys., 4
(1963), pp. 569-574.
[227] U. UHLHORN, Representation of symmetry transformations in quantum mechanics, Ark.
Fys., 23 (1963), pp. 307-340.
[228] B. L. VAN DER WAERDEN, The classification of simple Lie groups, Math. Z., 37 (1933),
pp. 446-462.
[229] V. S. VARADARAJAN, On the ring of invariant polynomials on a semisimple Lie algebra,
Amer. J. Math., 90(1968), pp. 308-317.
[230] O. VEBLEN AND J. H. C. WHITEHEAD, The Foundations of Differential Geometry, Cam-
bridge Tracts in Mathematics and Mathematical Physics, No. 29, Cambridge Uni-
versity Press, Cambridge, 1932.
[231] F. D. VELDKAMP, Representations of algebraic groups of type F+ in characteristic 2,
J. Algebra, 16 (1970), pp. 326-339.
[232] N. YA. VILENKIN, On the theory of associated spherical functions, Dokl. Akad. Nauk
SSSR, 111 (1956), pp. 742-744 (in Russian).
[233] , Bessel functions and representations of the group of Euclidean motions, Uspekhi
Mat. Nauk, 11 (1965), pp. 69-112 (in Russian).
[234] , Special Functions and the Theory of Group Representations, Amer. Math. Soc.
Mathematical Monograph Translations, No. 22, Amer. Math. Soc., Providence,
R.I., 1968.
[235] J. VON NEUMANN, The uniqueness of the Schrodinger operators, Ann. Math. Pura Appl.,
104 (1931), pp. 570-578 (in German).
[236] J. WEI, Note on the global validity of the Baker-Hausdorff and Magnus theorems, J. Mathe-
matical Phys., 4 (1963), pp. 1337-1341.
[237] L. WEISNER, Group-theoretic origin of certain generating functions, Pacific J. Math.,
5(1955), pp. 1033-1039.
[238] H. WEYL, The Concept of a Riemann Surface, first ed., B. G. Teubner, Leipzig, 1913
(in German).
158 Bibliography
[239] H. WEYL, Theory of representations of continuous semisimple groups by linear transforma-

tions. I, Math. Z., 23 (1925), pp. 271-309.
[240] , Theory of representations of continuous semisimple groups by linear transforma-
tions. II, Ibid., 24 (1926), pp. 328-376.
tions. Ill, Ibid., 24 (1926), pp. 377-395.
tions. IV, Ibid., 24 (1926), pp. 789-791.
[243] , Harmonics on homogeneous manifolds, Ann. of Math. (2), 35 (1934), pp. 486-499.
[244] , The Classical Groups: Their Invariants and Representations, 2nd ed., Princeton
University Press, Princeton, N.J., 1946.
[245] J. H. C. WHITEHEAD, On the decomposition of an infinitesimal group, Proc. Cambridge
Philos. Soc., 32(1936), pp. 229-237.
[246] E. WICHMANN, Note on the algebraic aspect of the integration of a system of ordinary
linear differential equations, J. Mathematical Phys., 2(1961), pp. 876-880.
[247] E. P. WIGNER, On the consequences of the symmetry of the nuclear Hamiltonian on the
spectroscopy of nuclei, Phys. Rev., 51 (1937), pp. 106-119.
[248] , The application of group theory to the special functions of mathematical physics,
Princeton University Lecture Notes, Princeton, N.J., 1955, unpublished.
[249] , Group Theory and its Application to the Quantum Mechanics of Atomic Spectra,
Vieweg and Son, Brunswick, Germany, 1931 (in German); English transl., Academic
Press, New York, 1959.
[250] L. S. WOLLENBERG, Sine and Poisson bracketsa distinction, J. London Math. Soc., 44
(1969), pp. 592-594.
[251] , Derivations of the Lie algebra of polynomials under Poisson bracket, Proc. Amer.
Math. Soc., 20 (1969), pp. 315-320.
[252] B. G. WYBOURNE, Spectroscopic Properties of Rare Earths, John Wiley, New York, 1965.
[253] C. N. YANG, Elementary Particles: A Short History of Some Discoveries in Atomic Physics,
Princeton University Press, Princeton, N.J., 1961.
[254] G. ZAMBOTTI, Factorization of differential equations and O(4) symmetry of the hydrogen
atom, Nuovo Cimento Suppl., 5(1967), pp. 954-962.
INDEX
A Basic weight, 93
Abelian group basis algorithm, 147 Bessel function, 72
Abelian ideal, 45 Bilinear form, 7
Abelian Lie algebra, 34, 44 antisymmetric, 8
Abelian Lie group, 34 matrix of, 8
Additive quantity, 142, 144 nonsingular, 9
Adjoint, 27 symmetric, 8
Adjoint representation, 52 Bilinear mapping, 40
Ado theorem, 53 Bilinear product, 12
Affine Lie algebra, 36, 44, 65 Boson, 142
Algebra, 12 Broken symmetry, 141
Algebraic group, 15
Analytic action, 20
Analytic function, 14 C
Analytic homomorphism, 32 Canonical coordinates, 26
Analytic manifold, 14 Cartan composition, 93
Analytic representation, 54 Cartan criterion, 53
Analytically related, 14 Cartan decomposition, 82
Angular momentum, 86, 97, 120 Cartan matrix, 73, 93
Anisotropic vector, 8 inverse of, 108
Annihilation operator, 30, 142 Cartan subalgebra, 66
Annihilator of module, 92 Casimir algebra, 98
Anticommutative product, 13 Casimir element, 72, 140
Anti-Hermitian matrix, 12 second order, 98
Antilinear operator, 27 Cayley numbers, 79
Antisymmetric bilinear form, 8 Character of module, 101, 102
Antisymmetrization, 60 computation, 110
Antisymmetrizer, 113, 125, 126 primitive, 113
Antiunitary operator, 27 Charge algebra, 143
Arc, 33 Charge multiplet, 144
Arcwise connected, 33 Chart, 14
Associative algebra, 12 Chevalley basis, 147
Atlas, 14 Circle group, 35
Automorphism of Lie algebra, 45 Classical mechanics, 24
Clebsch-Gordan coefficient, 137, 145-146
B Clebsch-Gordan series, 58, 102, 119, 120
Baker-Campbell-HausdorfF formula, 48, 52, Clifford algebra,
73 first, 132
Baryon number, 142 second, 133
Baryon octet, 144 Closed subgroup, 32
Basic module, 93 Coalgebra, 65
159
160 Index
Commutant algebra, 140 Determinant, 62

Commutation relation, 29, 86 Diagonal mapping, 65
Commutator, 12, 13 Differential equation, 20
Commutator of group elements, 17 Dimension of manifold, 18
Compact Lie group, 57, 78 Dimension of module, 102
Compact real form, 81 Dimensional analysis, 21
Compact real Lie algebra, 81 Diophantine linear equations, 147
Completely controllable, 47 Dirac correspondence principle, 28
Completely reducible module, 57 Dirac gamma algebra, 132
Complex analytic group, 78 Dirac quantization, 28
Complex general linear group, 6 Direct sum, 42
Complex orthogonal group, 6 Direct summand, 57
Complex special linear group, 6 Discrete Lie group, 33
Complex symplectic group, 10 Discrete space, 33
Complexification of real Lie algebra, 42, 81 Division ring of fractions, 64
Component of Lie group, 33 Dual basis, 41
Composite quantum system, 143 Dual module, 59
Comultiplication, 65 Dual space, 41
Configuration space, 25 Dynamical variable, 26
Conformal group, 21 Dynkin diagram, 76, 94
Congruence transformation, 9 Dynkin indices, 93
Conjugate elements of group, 101 Dynkin layer method, 107
Connected Abelian Lie group, 35
Connected space, 33
Conservation law, 26
E
Conserved charge, 142 Electric charge, 142
Conserved quantity, 26 Elementary module, 123
Constraint, 25 Elementary particle, 144
Continuous group, 19 Energy conservation, 26
Continuous homomorphism, 32 Energy level, 141
Contraction, 41, 59, 128 Engel theorem, 65
Contra variant index, 41 Equivalent representations, 55
Contravariant tensor algebra, 60 Euclidean group, 46, 71
Controllable system, 47 Exceptional simple Lie algebra, 77, 94
Coordinate neighborhood, 14 Exponential mapping, 11
Correspondence principle, 28 Extension of linear operator, 27
Coulomb excitation, 141 Exterior algebra, 60
Coulomb potential, 138 Exterior power, 61,123, 129-131,133
Covariant index, 41 External direct sum, 42
Coxeter diagram, 77 Extreme subspace, 92
Cross product, 13,24,45 Extreme vector, 91
Curvilinear coordinates, 25
Cyclic module, 92
F
Factorization method, 71
D Faithful, 53
Decomposable module, 56 Fermion, 143
Degenerate eigenvalue, 140 First Clifford algebra, 132
Degenerate perturbation, 141 Fitting null component, 65
Degrees of freedom, 26 Flow problem, 49
Dense subgroup, 46 Fok representation, 143
Derivation, 13 Fractions, 64
Derivation algebra, 13 Freudenthal algorithm, 110-111
Derived series of ideals, 44 Fundamental group, 34
Index 161
G Inversion operation, 7
Gamma matrix, 132 Invertible linear operator, 15
Carding domain, 29 Involutive automorphism, 82
Gelfand-Zetlin basis, 147 Irreducible module, 56
General linear group, 5, 11, 22 Isomorphic
Generalized charge, 142 modules, 55
Generalized coordinates, 25 vector spaces, 43
Generalized momentum, 25 Isotopic spin, 144
Generalized simultaneous eigenvector, 67 Isotropic vector, 8
Girdle, 110, 112 Isotropy subgroup, 47
Girdle division, 112, 119
Global structure of Lie group, 32-34 J
Goldie-Ore theorem, 64 Jacobi identity, 13
Graded algebra, 59 Jordan relations, 132
Grassmann algebra, 60
Group algebra, 112, 125
K
Kepler problem, 138
H Kernel, 32
Hadron, 144 Killing form 53, 104
Half sum of positive roots, 112 Kostant formula, 118
Hamiltonian, 25 Kronecker delta, 9
Hamilton's equations, 25
Harmonic function, 70 L
Harmonic oscillator, 30, 71 Ladder generating Lie subalgebra, 70
Helicity representation, 88 Ladder of roots, 73
Hermitian operator, 27 Ladder of weights, 70
Highest root, 109 Lagrange's equations, 25
Highest weight, 74 Lagrangian, 25
Hilbert space, 26 Laplacian operator, 71
Homogeneous Lorentz group, 88 Lattice, 42
Homomorphism of ideals, 44
of Lie algebras, 32 Law of energy conservation, 26
of Lie groups, 32 Layer, 107
of modules, 55, 137 Left-invariant vector field, 19
of vector spaces, 43 Legendre transformation, 25
Hydrogen atom, 140 Leibniz rule, 52, 58
Hyperbolic basis, 10 Length of root, 76
Level, 69
I Levi decomposition, 45
Ideal, 43, 60 Lexicographic ordering, 69, 108
Idempotent operator, 125 Lie algebra, 12-13
Identity component, 33 ideal of, 43
Improper orthogonal matrix, 6 linear, 53
Indecomposable module, 56 module over, 54
Indefinite metric, 10 nilpotent, 65
Infinitesimal group, 18 of Lie group, 15, 18,23
Inhomogeneous Lorentz group, 88 radical of, 45
Integer matrix inversion, 104 semisimple, 45
Integer-mode arithmetic, 103 simple, 45
Integrating factor, 20 solvable, 44
Internal direct sum, 43 structure constants of, 18
Invariant solutions, 20 structure of, 46
Invariant subspace, 56 subalgebra of, 43
162 Index
Lie group, 14 Module (continued)

Lie algebra of, 18 irreducible, 56
subgroup of, 31 quotient, 56
Lie module, 54 semisimple, 57
Lifting of bilinear mappings, 40 semispinor, 136
Light-like vector, 8 simple, 56
Linear form, 41 spinor, 135
Linear fractional transformation, 37 submodule of, 56
Linear Lie algebra, 53 tensor operator, 137
Linear mapping, 5 Module homomorphism, 55, 137
Linear operator, 5 Momentum operator, 86
antilinear, 27 Multiplicity of weight, 103
antiunitary, 27 Multiplicity problem, 147
extension of, 27
Hermitian, 27 N
in Hilbert space, 27 Natural pairing, 59
invertible, 5 Negative group, 126
matrix of, 5 Nilpotent Lie algebra, 65
orthogonal, 9 Nonassociative algebra, 12
self-adjoint, 27 Nonsingular bilinear form, 9
symplectic, 10 Norman-Wei solution, 47
unitary, 27 Nuclear physics, 11
Loop, 28 Nucleon, 144
Lorentz boost, 87 Null vector, 8
Lorentz group, 7, 10, 86-88
Lower central series of ideals, 65 O
Lowering algebra, 91 Observable in quantum mechanics, 29
Octet model, 144
Open subgroup, 32
M Orbit of group action, 20
Magnus solution, 47 Orbital angular momentum, 86
MaPcev Goto-Harish Chandra theorem, 45 Ordered basis, 69
Mass formula, 144 Orthogonal
Matrix, 5 group, 9
anti-Hermitian, 12 linear operator, 9
of a bilinear form, 8 matrix, 6
of a linear operator, 5 vectors, 8
orthogonal, 6 Orthonormal basis, 9
unitary, 6 Oscillator algebra, 143
Matrix Lie groups, 22 Outer multiplicity,-119
Matrix representations, 54, 55
equivalent, 55 P
Maximal Abelian subalgebra, 66, 72 Parent weight, 107
Measurement, 27 Parity, 7
Mixed tensor. 41 of permutation, 125
Module, 54 of Weyl group element, 112
basic, 93 Partition function, 110, 118
completely reducible, 57 Partition of integer, 125
cyclic, 92 Pattern recognition, 21
decomposable, 56 Pauli exclusion principle, 143
dual, 59 Pauli matrix, 36, 96
elementary, 123 Permutation group, 125
homomorphism of, 55, 137 Phase space, 25
indecomposable, 56 Physical symmetry, 28
Index 163
Physical system, 27 Root, 67

Plane Lorentz group, 81,85 level of, 69
Poincare-Birkhoff-Witt theorem, 64 positive, 69
Poincare group, 88 simple, 69
Poisson bracket, 26, 138 Root ladder, 73
Position operator, 88 Root space, 67
Positive definite, 10 Root system, 75
Positive group, 126 Root vector, 68
Positive roots, 69 Rotation group, 7, 23, 36, 46
sum of, 112 Runge-Lenz vector, 138, 140
Prime factorization for fractions, 115 r-vector, 61
Primitive character, 113
Projection operator, 125
Projective special linear group, 38 S
Proper orthogonal matrix, 6 Scalar control, 47
Scalar dynamical variable, 139
Scalar operator, 137
Q Schrodinger equation, 47
Quantization, 28 Schur's lemma, 137
Quantization axis, 83 Second Clifford algebra, 133
Quantum symmetry, 26 Second order Casimir element, 98
Quark representation, 100 Second quantization, 143
Quaternion algebra, 134, 136 Self-adjoint operator, 27
Quotient algebra, 60 Self-dual module, 95
Quotient Lie group, 32 Semisimple associative algebra, 125
Quotient module, 56 Semisimple Lie algebra, 45
Semisimple module, 57
Semisimplicity criterion, 53
R Semispinor module, 135
Racah coefficient, 137 Shifting rule, 70, 71
Racah formula, 119, 122 Signature, 10
Radical, 45 Simple Lie algebra, 45, 77, 78
Raising algebra, 91 Simple lowering element, 74
Rank, 66 Simple module, 56
Ray representation, 57 Simple raising element, 74
Real affine group, 35 Simple root, 69
Real form of complex Lie algebra, 81 Simple r-vector, 62
Real general linear group, 6, 33 Simple Weyl reflections, 77
Real orthogonal group, 6 Simply-connected, 33
Real restriction of complex Lie algebra, 42, 81 Simultaneous eigenvector, 67
Real simple Lie algebra, 81 Solvability criterion, 53
Real special linear group, 6 Solvable Lie algebra, 44
Real symplectic group, 10 Special complex orthogonal group, 6
Recursion relation, 71, 72 Special function, 70
Relativistic wave equation, 21 Special linear group, 5
Representation Special unitary group, 6
adjoint, 52 Spherical function, 70
equivalence, 55 Spin, 86,96, 115
faithful, 53 Spin and statistics, 142
matrix, 55 Spin group, 135
of a group, 51 Spin-half representation, 95, 96
of a Lie algebra, 51 Spinor module, 135
of a Lie group, 51 Spinor representation, 95
Representation space, 56 Splitting of Lie algebra, 45
164 Index
Standard Young tableau, 126 U

Stark effect, 141 Unimodular group, 6
State in quantum mechanics, 27 Unitary
Steinberg formula, 120, 121 group, 6
Strong interaction, 144 matrix, 6
Structure constant, 16, 18 operator, 27
Subalgebra, 43 representation, 57
Submodule, 56 Unitary Symplectic group, 11
Subspaces, Universal covering group, 34
lattice of, 42 Universal enveloping algebra, 64
sum of, 42
supplements of, 43 V
Superselection rule, 28 Vector dynamical variable, 139
Supplement of subspace, 43 Vector field, 18,49
Symmetric bilinear form, 8 Vector group, 34
Symmetric space, 5, 71 Vector part of real form, 83
Symmetrized tensor, 128 Visual perception, 21
Symmetrizer, 126
Symplectic group, 7, 10 W
Symplectic linear operator, 10 Weak external direct sum, 60
Wedderburn theory, 125
T Weight, 66
Tangent bundle manifold, 25 Weight ladder, 70
Tangent space, 17, 18 Weight submodule, 67
Tensor, 39 Weight system, 66
Tensor algebra, 60 calculation of, 106
Tensor analysis, 123 Weight vector, 67
Tensor operator 137 Weyl formula
Tensor operator module, 137 for Casimir element, 98
Tensor power, 59, 124 for characters, 112
Tensor product, 39 for dimensions, 115
of a linear mapping, 40 Weyl group, 75, 77, 121
of a vector space, 39 storage of, 110
of vectors, 39 Weyl reflection, 75
Tensoring together, 60 Wichman decomposition, 48
Topolqgical group, 14 Wigner-Eckart theorem, 137
Toroidal part of real form, 83 Wigner theorem, 28,57
Torus subgroup, 101
Totally disconnected space, 33
Trace, 8 Y
Young shape, 125
Trace removal, 129
Young tableau, 126
Traceless matrices, 12
Trajectory, 25
Transposed matrix, 10 Z
Two-dimensional Lie groups, 36 Zeeman effect, 141
Two-sided ideal, 60 Zero-girdle, 112

Johan G. F. Belinfante, Bernard Kolman

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Johan G. F. Belinfante, Bernard Kolman

Hochgeladen von

Copyright:

Verfügbare Formate

A Survey of lie Groups

and Lie Algebras

Classics in Applied Mathematics

*'First time in print

Library of Congress Cataloging-in-Pdblication Data

Includes bibliographical references.

SL3J1L is a registered trademark.

We express our thanks to Harvey A. Smith who collaborated with us

The theory of Lie groups and Lie algebras is an area of mathematics in

Since much of the applied literature nowadays presumes a quite detailed

LIE GROUPS AND LIE ALGEBRAS

1.1 THE GENERAL LINEAR GROUP

where and A is a scalar.

1.2 ORTHOGONAL AND UNITARY GROUPS

Other important subgroups of the general linear group GL(n, C) are

The special unitary group is

The determinant of an orthogonal matrix can only be + 1 or 1. This is

The geometrical interpretation of the distinction between proper and

1.3 GROUPS IN GEOMETRY

Matrix Lie groups also arise in various types of non-Euclidean geometry,

A bilinear form (x, y) can be completely described by its matrix /?,-_,-

A bilinear form (x, y) is symmetric if (x, y) = (y, x) for all x and y in V.

We can make use of such congruence transformations to bring the matrix ft

where the Kronecker delta <5i; is equal to 1 if i = j and 0 if i ^ j. The matrix of

for an antisymmetric nonsingular bilinear form. Nonsingular antisymmetric

for all 1,7 = ! , - , .

where 1 denotes the n x n identity matrix. The matrix group Sp(n, C) is

where AT denotes the transpose of the matrix A. Closely related to this

We may also mention here the unitary symplectic group,

1.4 THE EXPONENTIAL MAPPING

where 7 is the identity matrix, and each matrix element converges

1.5 LIE AND ASSOCIATIVE ALGEBRAS

Finally we assume that the Jacobi identity

1.6 LIE GROUPS

coordinates of z be analytic functions of those of x and y. We need not

1.7 LIE ALGEBRAS OF LIE GROUPS

If the coordinate neighborhood is taken sufficiently small, the terms indicated

where Slj is the Kronecker delta and

A similar argument can be given by setting x equal to the identity, leaving y

It seems therefore that the local structure of a Lie group is determined by

In these coordinates, group multiplication is given locally by

For convenience we may assume that such a coordinate transformation has

It is of interest to study what group operation corresponds to this

Solving for (x~ *)'. we obtain by repeated substitution,

The commutator of two group elements x and y is the element

The computations given previously can now be interpreted as operations on

We can regard this as a form of vector multiplication and introduce the

1.8 VECTOR FIELDS

We can obtain another picture of the geometrical meaning of the Lie

Since U Vg = V Vg = 0, it follows that W Vg = 0, and hence W(x) is

1.9 LIE THEORY OF ONE-PAR A METER GROUPS

The abstract definition of a Lie group as a manifold was an outgrowth

Suppose we are presented with a first order differential equation

with analytic coefficients. There exists a one-dimensional Lie group of

with the group and the operator

While every first order differential equation can be shown to be invariant

1.10 MATRIX LIE GROUPS

of the technicalities of manifold theory [73]. However, this would definitely

The product of two matrices in these coordinates is given by

The tangent space has for a basis the matrices

about the axis For any angle co ^ 0, we have

where eijk is + 1 (- 1) if (/,;, k) is an even (odd) permutation of (1, 2, 3) and 0

Then an element of the Lie algebra so(3, R) is an antisymmetric matrix